4’

 

 

rail?“ .Fxphffifluvql

ildiU ' ’llU

-'-. .1_~'~‘:A ,'!’
ran-mange (-6“: n. {31.5- .
V ”war ’1‘ _.
ohm-”Lube: «ma; @431in- 7‘

Research Method and Procedure

i—lll’l’l 5' o o ' '
' m Agncultural Economlcs. -

‘5 I _' ‘L.’ l
[’4 /. _. . I,
.395“ A Publication of the Adviso Committee
,2 ry
,5 .
., i—H, . on
3M . I! l/
,1 4-
':\"~' '. . .
,1} ‘ 1 r Economlcs and Socxal Research
in
Agriculture
of

The‘kaocial Science Research Council

VOLUME ONE

AUGUST 1928

\—

; m mterial in this volume Was preparad under the
direction of the Advisory Committee on Economic and Social
Research in Agriculnuro of the Social Science Research
Council. The Constituent organizations of the Social
Science Research Council are as follows:

American Economic Association
Altar-loan PolitiCal Science Association
41.1118:ij Sociological Sociocy
American Statistical Association _
American Psychological Ass: :cs ration
American Anthronological Ins sociation

‘ American Historical Association

The members of the Advisory Oomdttee on Social and
Economic Research in Agriculture are:

H. C. Taylor, Northwestern University, Chairman
John D, Black; Harvard {Taiwan-icy

Joseph S. Davis, Food Research Instituto

C. J . Galpin, Bureau of Aggioulfiurul mommies
L. 0. Gray, Bureau of Angcfitulel Ecommids

E. G. Hearse, Institute of Economics

G. 3'. Warren, Cornell University

kmm'hém .ef‘ Maw IWsow Gomfioa

on

Ecoziomioa and Social Research

in

Agriculture
of

a.

Thaw wince Roamc’h 0mm

 

 

v us»

1;...

 

   

 

 

     

T.):~‘.‘a.2

3;... A“?! u; m w\

-“

mm 2.... i '-

 

)I‘AH

L“ ’
,3 b:
,. .> (I ‘ , _

V."

if] \ lf-(Z’

The special sub-committee which presents this body of material
of research methods in agricultural economics is supremely conscious of
its shortcomings. Except for two months devoted to it by the chairman
of the sub-committee, it has all been contributed out of the busy lives
of those whose names you will find attached to the various sections of
it, and a considerable list of others who have helped in ways less easy
to acknowledge. There are important gaps in it. In some eases these
hays resulted because those assigned certain sections were unable to
finish them in time. Smaller omissions have resulted because the mem-
bers of the sub-committee did not think of them early enough. A good
many other gaps have been filled by the members of the sub—committee.

PREFACE

Limitations of space hare compelled almost ruthless conden-
sation of many of the contributions. For example, the section on "The
Measurement of Time Series Movements" was condensed from 150 pages of
manuscript submitted. The equivalent of 120.pages of mimeographed
material and a score of charts were deleted in e second cutting in order
to keep within the publication budget. ”

The practice is followed of creditingféhchycontributor in the
text or in a footnote. This does not mean that he is altogether res~
ponsible for what follows, beCause the subécommittee has made occasional
changes in the interest of consistency. Nor does the sub-committee
wish to be held responsible for all of it, since it has in places ac-
cepted statements by contributors which it would not accept as its own.
The sections which are not credited to any one person are mostly the
work of the cheirman.

If this publication was the product of the sub—committee, its
members would wish to express their appreciation of the splendid as—
sistance which they have had from a large group of co—workers. Inas—
much as it has been a cooperative venture, a more appropriate statement
is in terms of the pleasure and benefit which they have derived from
the joint effort. _

John D. Black, Chairman
' L. C. Gray

_E. G. Nourse

H. R. Tolley

kugust 13, 1928.

"O
V ,\)r‘~4

TABLE OF CONTENTS

Introduction, p l. *
What Is Research? p 3.
What Is Science? p 6.
Scientific Method. p 9.
Qualitative vs. Quantitative. p 11.
Inductive vs. Deductivc. p 15.
The Scientific Methods. p 15.
Social Science Research. p 18.
Economics vs. Social Science Research. p 19.

Part I Research Procedure. p 21.
Choice of Project. p 21.
Planning the Project. p 26.
Execution of the Project. p 82.

Part II Methods. p 36
A. Statistical Method. p 36.
(a) What 18 Statistical Method? p 36
(b) Units and Measures - Definition of Terms. p 38.
(c) Errors in Data. p 47. .
((1) Sampling. 1) 49.
(e) Soucces of Ddhe- and methods of Securing It. p 58.
(1) The Survey Method. p 58.
(2) Dita of PriVate and Public Records. p 91.
(3) Supervised Records. p 104.
(4) Mailed Questionnaires. p 113.
(5) Date from Reporters. p 127.
(6) Secondary Data. p 138.
(f) Preparing Data for Analysis. p 140.
(1) Editing. p 140.
(2) Checking for Accuracy. p 144.
(3) Tabulating; p 144.
(g) Analysis of Data. p 146.
(1) Summarizing. p 146.
Frequency Distribution. p 148.
Averages. p 152.
‘ Diapersions. p 156.
(2) Measurement of Time—Series Movements. p 159.
Types of Movements. p 159
Secular Trend. p 162.
Seasonal Variations. p 182.
CycliCal Movements. p 188.
Periodicity. p 189.
Lag. p 193.

 

RESEARCH METHOD ”sweetness
L11 AGRICULTURAL scone-Mics- ‘

INTRODUCTION

_ y The objective.of the Committee on Social and Economic Research
in Agriculture in preparing this volume is to summarize as much as pos-'
sible of the eXperience of agricultural economists in planning, organ-
izing and executing research in this field.’ This handbook is not in-
tended to be a textbook on research method. Formal research method as
such is much more completely developed in special treatises devoted to
it. But the application of formal method to any problem brings to light
numerous details and special considerations. It is these latter which
constitute the material of this handbook. ‘ -

The need for research in agricultural eeenbmics needs little
emphasis here. 'From the beginning, teachers in this field have felt
keenly the need of more facts as to the economic aspects of farming
and marketing activities, and there has been a growing realization of
the need for understanding the.relationships involved. Economic facts
are transitory. Students who are taught agriCultural economics in terms
of specific rule—of-thumb practices or economic set—ups may find them-
selves out-of-date in a few years. What they need is training in econ-
omic analysis and in the.use of data, especially the data of changing
,economicrconditions. ' ‘ ., ' “ '

That extension workers have come to realize the importance of
. economic research needs no ampler teetimony than that which‘follows,

- taken from a report made on April ll at the Joint Meeting of Directors
”of Extension with officials of the Bureau of Agricultural Economics and

the Extension Service of the U; S. Departmentjof Agriculturezl

"There is need for a strong research program
in economics in the respective states. ‘

"The Bureau of Agricultural Economics of the
Department of Agriculture can render valuable ser-
vice by studying, upon request, the research pro-
grams and teaching curricula in the'respeetive
states and assisting in coordinating and broadening
the work in this field.

"Inasmuch as the last session of the Land Grant
College Association was devoted to a consideration
of the agricultural problem, its causes and remedies,

- it is recommended that the seSsibn this year deal .
with strengthening the research, teaching-and exten-
sion progrmns in the economiC-phases of agriculture,
especially as referred to in this report."

. The recent protracted struggle for farm relief legislation has thus
far had no more positive result than making the leaders in this movementrj
' and in national affairs realize that better and more'complete data ad-
equately analyzed is basic to any program of orderly production or order-
ly marketing, and to the success of any administrative device set up'to
help agriculture with its manifold problems of production, land-utilise
ation, marketing, credit, insurance, transportation and taxation. The
.passage of the Purnell .Act is itself testimony to this fifi§0§.- The
ninth recommendation of the Business Men‘s Commissibn on Agriculture is
,as follows: .- ‘7“ ' '
"The Commission strongly urges the extension‘
and coordination of research work in the field of
agriculture by the Federal and state governments
and other agencies and the appropriation of larger
funds for such work. Extensive research is needed.
to supply the basis of a comprehensive land'util-
ization policy, for the elimination of plant pests
and diseases, for the development of new types of
agricultural products and of new uses for existing
products, as well as concerning the possibilities
of the application of industrial methods and busi-
ness organization to agriculture, all of which
agriculture itself is far less well equipped to
further than are other fields of economic activity.

LL:

)
Finally, the Budget_Director of the U. S. Government has come to the.
point of including in his category of "necessary expenditures" some ad-
ditional appropriations for the U. S. Bureau of Agricultural Economics.

With all this growing emphasis upon research, no argument should be-'
necessary as to the impertance of research method. But the whble weight
of circumstances is against giving sufficient attention to methodology. 7
'The administrative authorities want immediate results. The research . .
workers themselves look anxiously forward to specific results. iii the‘
results accord with expectations, the method is likely to be accepted
as.sound without further testing. New projects outlined take orer '
methods already in use without seriously questioning their validity. The
Vmadority of economists and even of research workers themselves are ins
patient with discussions of research. "The problem suggests the appro-
priate method", is their answer to questions asked on this subject; or
"Good research workers are born not trained"; or "Good sense is a suf-
ficient guide". ‘

3

But many signs point to a change of attitude toward research method.
'Directors of state experiment stations have recently become more critical
of research methods in agricultural economics. They have too often seen
.the old methods produce doubtful or inconsequential conclusions.l As long
as research in agricultural economics was an insignificant part of their
program, they left it largely in the hands of their economists. Now
that the passage of the Purnell Act has made it the new direction of
growth_of their activities, many of them have for the.time given the
major part of their thinking to it. These men were mostly trained in
the natural sciences. They bring to the analysis of research methods

3

in economics the point of view and the standards of the natural, sciences;
and their reactions have constituted a challenge to research in this field
which has been and which will continue to be of tremendous value to it.

No less valuable is the service.of similar nature rendered by Dr. Allen

of the Office of Experiment Stations. His analysis of the new research
projects submitted under the Purnell Act has been a strong influence in

the right direction. The economists themselves-have had more critics of
research methods withing their own ranks. The result has been that any
method proposed for a problem is likely to be challenged today; under these
circumstances the preporent of it feels called upon to defend it. This ,
starts him to thinking vigorously about it. In this way, routine procedures
in‘research have been brOken down and an era of growth has set in. Projects
are being set up increasingly in which development of method is either the
sole or an important objective. A contribution to method, it is coming to
be realized, is likely to have many times the usefulness of a contribution
to the analysis of a problem, particularly of a problem in a given area.
Many of the younger workers are bringing fresh points of view to the depart-
ments to which they come. -

Entirely aside from questions of research method themselves are those
relating to the organization and administration of research. 'Certain
large aspects of the organization of research.in agricultural economics
have recently become important, such as the relationships between the
research work of public agencies and institutions of research, between
that of publicly supported and privagely endowed institutions, that of
state experiment stations and the U. S. Department of Agriculture, and
that of state experiment stations and state departments of agriculture _‘
and markets. All this points to.the need of the coordination of re-
search in agricultural economics into something approximating a national
program. The final section of this handbook will be devoted to matters
of the latter sort. ' - ‘

What i_ At the outset, we shall do well to have as clear'
Research? a notion as possible as to what research is and the
nature of scientific method. It is doubtful if we can
set up final criterion by which we can definitely class-
-, ify certain things as research and others as not. All
will agree that more fact-finding, or mere census taking, or mere making of
records, is not research. But clearly the securing of accurate data or
records as in a survey is part of the procedure of research in the social
sciences. It is this whether in the organization of research one person or
agency gathers the data or source material and also analyzes the data, or
under a system of division of labor, one party or agency collects the data
and another analyzes it. To be sure, a project which has for its sole ob-
ject and final and the assembling of the data cannot of itself be called
a research project; but if it can.properly be said that the data have been
collected in such a way, and in the expectation, that they will presently
be analyzed by others, or used in related research projects, then such a
project can surely be considered a part of a research program. It is per-
tinent to remark, however, that when the fact-gathering and analysis are
not definitely tied together in one project, or are done by different
agencies far removed from each other, then fact-gathering is in great
danger of being mechanical or routine.

 

Analysis itself is of different degrees of research value. The
mere computing of averages, for example, has little research signifi-
cance. Analysis which looks to more description of an area, or group
of marketing agencies, or an historical period, is frequently research
of a low order if at all. If the description-is so developed-as to
show relationships between variations as well as the variations them-
selves, or similarly, relationships between the different changes occur-
ing during a period, then it becomes research_of a much higher order.

If in addition, these relationships are so analyzed that they can be used
as a basis for inferences of more general application, the research value
of the research project is still more apparent. ‘- -

A more.difficult question is whether an analysis that looks to pro-
viding facts and conclusions needed for a program of amelioration is
pneperly called research - let us sa , a study of farm organizations in
an‘area looking forward to an extension program based on the conclusions
reached. Undoubtedly such projects stand as research in the minds of some
experiment station directors. They have as much right to be classed as
research as projects developing methods of control cf insect posts or
rations for fattening steers. But the devotee of "pure" science is like-
ly to look askance at such ventures and ask if there are no general prin-
ciples relating to combinations of enterprises the understanding of which
cannot be advanced a little by such a project. she-expects an experiment
in feeding to reveal some relationships of general or more far reaching
nature between amounts and proportions of different feeds and gains in
weight. There is undoubted validity in this point of view; but it may
be carried altogether too far. Results of most economic analysis in
general have a more local and more temporary application than those-in
the applied natural sciences. This must not be taken as evidence against
their essential research nature. It is the inevitable consequence of the
essential characteristic of social science research, namely, many inter-
related factors affecting the result. Always a good many of these factors
are localized or placed in time. This does not mean that there are no
factors that extend over large areas, or continue from period to period;
but rather that the task of isolating and measuring them is made more
difficult. Indeed, the whole task of research is made the more difficult,
especially that part of it involving inference. '

On the other hand, a project designed to serve a purely local or a
purely temporary need, without contributing anything of more general or
permanent value, especially the latter, ordinarily can have little scien-
tific value., To call it research would be somewhat akin to calling re-
search the task of the social worker in locating a starving family in
order that it may be properly fed, or locating a supply of apples to
overcome a temporary shortage in a market. ‘ " d

The point of view of the scientist himself on this question has been
as well expressed by Karl Pearson as anybody:

"I want the reader to appreciate clearly that
science justifies itself in its methods, quite a-
art from any serviceable knowledge it may convey.
We are too apt to forget that this purely eduoa.
tienal side of science in the great value of its

practical applications. we see too often the plea
«usraisedrfor science that it is y§3§21LEQQEl§§gi.
hiwhile philology a“- L ,‘and philosoyhy are _,‘
supposed to have small utilitarian or commercial
value. Science, indeed, often teaches us facts of
primary importance for practical life; yet not on
, this account, but because it leads us tp classifi-
jcations and systems independent of the individual
thinker, to consequences and laws admitting play; '
‘room for individual fancy, must we rate the train—
. ing of science and its social vabie higher than
those of philology and philosophy. Herein lies
the first, but of course not the sole, ground for
,the popularization‘of science. That form of pop-
ular science which merely recites the results of
”investigations, which merely Communicates useful
knowledge, is from this standpoint bad science,
‘ or no science'at all."*' ' ‘

' The characteristic of economic research which the natural scientists
have most,difficulty in evaluating is its dynamic aspect. They are likné‘
1y to say of a project that if the facts and relationships it discloses .
are not likely to be true a few years from now, why discover them? This '
Would be a sufficient answer, looking at the matter from a purely research
point of view, if such facts and relationships were to be obtained only *
once and at this time. But if this is the beginning of a series of such
analyses made annually or at other stated intervals, looking to the ob«
servation of trends and discovering of relationships between variables, ,
in time sequence, then the answer would be altogether wrong. The-close ‘
analysis of year—to-year and month—tOumonth changes in potato prices
from 393 pp is research'of the same order as analysis of the factors
affecting the price of potatoes 13 ygags past. hesearch.basic to outlook ‘
analysis can be research of a high order if it is.planned so as to pro-
vide the basis for a continuing analysis of economic changes. Economic
analysis of any problem is never done. Setting up adequate machineny '
for collecting data of current changes, as of land values or shifts in
crops, so as to make possible continued testing of hypotheses, may pro--

3 perly be considered a part of the research project itself the same as

the collecting of survey data. Of course the purpdse for which the data
are collected is important. If it is merely to enable someone to pre-
dict.the size of the coming lamb crop or its price, end'to tell the fare
mer that to do under the circumstances, then the collecting of data, is
a part of service work and not research. ' '

‘It is in the sphere of dynamics that economic science has been most
slow to develop. The laws and principles of economic change offer a rim»

n.’ J.

' .field for the social scientist. The close study of current phenomena 6

probably has more to contribute to progreSs along this line than the am-
alysis of the records of the past. DeveiOping the technique.of obser-
vation.and measurement of current changes is one of the important rese-
arch needs of the day. ,> ' I ‘

All

it.

* Grammar of science, p. 9;

We shall have a better comprehension of what‘fl

:fiha§.;§ constitutes research if we consider for a moment
Science? the nature of science itself and scientific method.

' Karl Pearson says in his Grammar of geigngg that
"all knowledge is concise description". He is here

thinking of the knowledge which constitutes science, not of the isolated
fact. Scientific knowledge consists of statements, in.unmistakable lan—
guage, of the uniformities and regularities in the phenomena of the
universe. Once such a statement of a uniformity or regularity is made.
it largely saves the necessity of special examination of the pertinent '/
aspects of individual cases that arise. It becomes possible to reason
from these general statenents to statements as to the particular case.
This is the economy of science which has made human progress possible.
- "In the struggle for existence," says Pearson, "man has won his dicta-
torship over other forms of life by his power of forseeing the effects
*which flow from antecedent causes - not only by his memory of percep-
tions— --------- science is an intellectual resume of past experience,
and mental balancing of the probability of future eangierience.“I

One of the principal reasons that only probabilities can be stated
is that the new causes are never the same as the old causes. To quote
from Pearson further:

"Everything in the universe occurs but once; there

is no absolute sameness of repetition. Individual

phenomena can only be classified, and our problem

turns on how far a group or class of like, but not
. absolutely same, things which we term "causes",

‘ will be accompanied by or followed by another group
or class of like, but not absolutely sane, things
which we term "effects" ------ "If the "causes have

.such and such a degree of likeness, how like will
the "effects" be?“**

"Existences are individual; it is a human, a
rational process which for economy of thought clas—
sifies them. Any variation within the existences
in one class is bound to be associated with a cor-
responding'variation among the existences in a
second class. Science has to measure the degree of
stringency,.or of looseness, in these concomitant
variations. Absolute independence is the concep-.
tual limit at one end to the loosenese of the link;
absolute dependence is the conceptual limit at the
other end to the stringency of the link. The old
View of cause and effect tried to subsume the univ—
:erse under these two conceptual limits to experie
ence -- and it could only fail: things are not in
our experience either independent or causative.

All classes of phenomena are linked together, and

 

v 0p”; oit., p. 113.
t* "g " " 157.

While there is no such a thing as absolute causation in the universe.-

the problem in each case is how close is the degree
of association. Likeness of Causes produces like-

ness of effects; we can measure the degree of like-
ness, whether we are dealing with a chemical react-
ion or with the resemblance in any aptitude between
parent and child. There is no question of absolute

' sameness in either case; there is a wide degree of

difference in the likeness, but both problems are
only variants of one and the same logical problem”-

'the contingency problem at the basis of modern

science.*

"No phenomena are causal; all phenomena are
contingent, and the problem before us is to measure
the ‘. degree of this contingency, which we have
seen lies between the zero of independence and the
unity of causation. That is briefly the wider out-
look we must now take of the universe as we exper-
ience it."""il -

.but only probability, nevertheless there is such a large amount of uni-
formity and regularity that it becomes possible to build.up a large
structure of laws and principles and thus introduce some very great

economies of intellectual effort in human affairs.
economies of this sort prove to be false economy.

Some attempts at
The statements of

regularities or uniformities first take rank as hypotheses or theories.

Eventually-many are accorded the ranking of laws or principles.

There

is no infallible guide as to when a proposition is entitled to be gen-
erally accepted. The test of validity is acceptance by the great pre-
ponderence of "normally constituted and duly instructed minds."***

There can be no test outside of the human mind.

While there is no necessity in the world of experience - even the
sun may conceivably not riSe tomorrow - there is in the world of logic

. and concepts, if we accept the canons of Aristotelian logic. If we ‘

state our assumptions and premises in a certain way, or set up certain'

concepts, there is no escaping the conclusions. But the conclusions are

only as true as the premises or the concepts.'

Sound logic is as fundamental a part of good science as are accurate

data and facts. "The word science", says Pearson, "applies to all rea~

soning about facts which proceeds from their accurate classification, to

the appreciation of their relationship and sequence."****v'

It will be clear from the foregoing that classification is a highly
-important part of science. It is a necessary condition to the attainment

of that~conciseness of description which is the essence of science.

is necessary to any comparing of causes as a basis for forecasting effects.

It

 

t Op. cit., p. '155 m- 0p. cit., 13.3314

'. is n n n 174 '**m* u u

/

n-~~ ~. _

-~‘s. The common statement that "science is classified knowledge" will not suffice;
“however, unless understood to include also the laws that follow from this
classification. ~

; Karl Pearson's definition of classification isimportant-here: "The
reader must be careful to recollect that classificatign is not identical
with collection. It denotes the systematic association of kindred facts,
the collection, not of all, but of relevant and crucial facts;";

The statement that a scientific law is merely'a description may be
easily misunderstood to mean that science is only empirical in the usual
sense of that term; that is, as consisting merely of simple statements of
association between things. The analytical method, which is the true meth—
od of science, is not satisfied with simple statement of association, but
attempts to get back of this to other associations and back of this to
§t111 other associations. To illustrate, an empirical analysis of elastic- ~

-ity of supply of milk would stop with discovering what factors produce
change in the supply of milk, and perhaps measuring- the amount and time ~
of the response. A scientific analysis would reveal the relationships on
the dairy farms which led to these responses. and their amount and time.»
The analysis might be pushed still farther into the social psychology_ of '
the group of farmers and other more ultimate factors. But as Pearson so
well remarks, scienCe has no first cause. A cause. as understood in this
connection, is one stage in the routine ”of experience. Back of this stage,
are others which science is not now able to compr(hend Pearson observed
that the problem of the causes of the growth of a particular ash-tree in
his garden could be widened out into a description of all the past stages
of the universe. A scientific analysis is just a few links in a long chain.~
of cause and effect. The essential difference between empirical and analy-
tical is that the one accepts superficial relationship without inquiring
as to antecedents, whereas the latter pursues antecedents a stage or two
at the least.

The importance of the analytical method is that it enables us to

forecast with greater assurance. If we know what lies behind a given set

of relationships, we are in a much better way of knowing whether they will
occur again. A study of the factors affecting the yield of cotton dis-
closed an apparent association between winter weather conditions and yield
in the subsequent season. What lay behind this was not detirmined. When
a longer series of data became'available, the apparent aesociation of the
period appeared to be largely chance. If a reason had been established for
this association of winter weather and yields. the association would not
have proved to be chance. and a forecast of yields could have been made..

with same assurance even on the basis of data for a relatively shorter
per 0

Professor E. B. Wilson of the Harvard Medical School, in his paper
on the use of quantitative method in economics before the American Econ-
omic Association in December 1927, wrote as follows:

"Even if progress be slow, the very effort _at
rationalization is itself a stimulus, and abandoning
the effort leads to a decay or relaxation of science

‘01) cit p.77 . -‘

even in its empirical parts. You are doubtless a;
were that around 300 3.0. great beginnings were
made in medical science under Herophilas and Era»
istratus, who avoided all dogmatism, reliinggon;
careful observation and analysis, but that aftere
wards came the school called Empiric which con-
fined itself to purely descriptive work and pro—
hibited the inquiry into the general causes of
things. Progress stopped.*

"It goes without saying that both (analytic-'
a1 and empirical) methods may be mutually helpful
and that both are employed together, albeit with
differing emphasis according to the nature of the
phenomenon or the individual proclivities of the
investigator. The aim of the empirical nethod is

_to enable us to recognize a situation in such a

. manner that we may distinguish between different
situations. This is basic to knowledge; we cannot
analyze and reason until we haye some more or less
definite information and some fairly well discrim-
inated notions which we may analyze and upon which
we may reason. The aim of the rational method is

to any crgg, to prove, to forecast with greater
assurance."**

We can now discuss effectively the essentials of

.Scientific scientific method. A few statements from Pearson‘s
Method Grammar 9: Science contains most of what we need:

"As to the scientific method, we saw in
our first chapter that it consists in the careful
and often laborious classification of facts, in the

H comparison of their relationships amd sequences,
and finally in the discovery by aid of the disci-
plined imagination of a brief statement or formula,
which in a few words resumes a wide range of facts.
Such a formula, we have seen, is termed a scientific
law. The object served by the discovery of such
laws is the economy of thought; the suitable assoc-
iation of cones tions drawn from stored sense~
impressions, pegmits the fitting exertion to follow
with the minimum-of thought upon the receipt of an
immediate sense—impression. The knowledge of
scientific law enables us to replace or supplement
mechanical association, or instinct, by mental as-
sociation, or thought. It is the forethought, by
aid of which man in a far higher degree than other
animals is able to make the fitting exertion on the
receipt of a novel group of sense-impressions."***

*gg 0p. cit. p. 2-
n.’ I: II n ‘1

*t* n I ‘II'77

 

10

-“
.114

"The classification of facts and the formation
of absolute judgments upon the basis of this class-
ification-- judgments independent of the idiosyncrasies
of the individual mind-~essentially sum up the gap and
method of modern science. The scientific man has above
all things to strive at self-elimination in his judgments,
to provide an argument which is as true for each individual
mind as for his own. The classification 21_fa¢ts. the
recognition of their sequence and relative significance is
the function of science, and the habit of fonming a judgment
upon these facts unbiassed by personal feeling is character-
istic of what may be termed the scientific frame of mind."'

"The scientific method is marked by the following
features:--(a) Gareful and accurate classification of facts
and observation of their correlation and sequence; (b) the
discovery of scientific laws by aid of the creative imigina-
tion; (c) Self-criticism and the final touchstone of equal
validity for all normally constituted minds."“

Also the following from a paper by J. C. Cobb entitled "The

Significance and Use of Data in the Social Sciences" in the Economic
Juurnal of England, March 1928:

"The scientific method assumes a clearly stated
problem, analyzed to make sure that it is a simple prob-
lem and not several problems combined. It requires a
careful study of factors bearing on the problem and
quantitative statement of each. A problem so stated
is in form to have logical reasoning produce a solu-
tion. The reasoning must of necessity always remain
dependent on the clearness and soundness of the reason-
ing mind, but if the problem is fully and simply stated
it is comparatively easy to follow and check up the
logic of the reasoning. The great difficulty in the
scientific method is the creation and verification of
quantitrtive data."***

"It may therefore be stated that the purpose
of scientific investigation in social science is the
production of data which logical reasoning can apply
to and use in the conduct of human affairs."*"’

From the foregoing the following can be summarized as the

essentials of good scientific method:

1: Careful logical analysis of the problem, isolating it from
other problems and separating out its elements. This
means, in some cases, the formulation of an hypothesis,
in the proper meaning of this expression - a trial
hypothesis that will point the investigation.

 

’0‘

013. 0113.13. 76.
Economic Journal of England, Ibrph 1928, p. 73.

II I! 0' N W 68 .

‘-.V.

11

2; Unequivocal definition of terms and concepts, and statis-
tical units and measures, so that others Will under~
stand exactly, and be able to repeat the analysis and
test the generalizations. I

3: Collection of'cases and data pertinent to the subject in
hand. I M

4 .

4: GlasSification of cases and phenomena and data of same.

5: Expression of factors in quantitative tenns whenever pos-
sible. ‘ ‘ '

6: Rigorous and exacting experimental or statistical proce-
dure in summarizing the data and in isolating the at-
tributes or variables and measuring their relationships
and'inter-effects.

7: Statement in unassailable terms of the exact conclusions
that are warranted for the cases examined.

8: Sound logical reasoning as to the bearing of these con-
‘ clusions on the trial hypothesis and in the formula—
tion of generalizations,

9: Statement of cbnclusions or generalizations definitely
and clearly so that others will be able to check thenn

10: Complete elimination of the personal equation.

11: Complete and careful reporting of the data and the methods
of analysis so that others can check the analysis, or
test the generalizations with new sets of data.

Perhaps it is well at this point to repeat that the unassailable
conclusions referred to are not statements in terms of absolute cause
and effect, but of degrees of contingency and probability. The con—
clusion may even be in terms of negative results, or of a small de-
gree of probable contingency; ~What is wanted is statements that are
amply supported by the data and analysis. ' "

Quantitative gs ‘ In the foregoing list of essentials of good
.anlitative scientific method, the demand appears for both
quantitative and.qualitative analysis, perhaps in
about eQual proportions. There has been consider-
able discussion recently as to the relative import-
ance of these two types of analyses(x). _Most of such discussion is
futile. Both methods are'necessary. In some fields, only qualitative

 

(x) See Proceedings of the American Economic Association 1925, for
Professor Mitchell's preSidential address, and the proceedings of‘;
the two following years for round tables on methodology. - .-

‘“enalysis has thus far been possible. No one denies that-analysis should,
be'reduced to quantitative terms whenever possible. Most important of
all to realize is that even When quantitative data are available, a
large amount of qualitative analysis is needed. It is needed at the
very start in isolating the problem and separating out its elements.

_ in stating the variables, in defining the units, concepts and meas-
ures, and in selecting thedata to be collected. It is even more

‘\essential in formulating the conclusions 1

Mr. J. C. Cobb, in the article previously referred to dis-
tinguishes between the terms qualitative and quantitative as applied
to data in the following language:

"Qualitative data are factors, elements or con-
in ditions observed or ascertained to exist in phen—
omena, the relative value or comparative importance
of which have not been measurably stated and de-
fined.

"Quantitative data are such factors, elements
or conditions which have been subjected to inten-
sive analysis and their relative or comparative
force or effect on a phenomenon defined, measured
and stated.

_ "It will be noted that under these definitions
qualitative data are necessarily the precursors of
quantitative data, as the existence of a factor
must be observed or ascertained before its force
and effect can be measurably considered." (X)

It follows from the foregoing that qualitative analysis is
analysis in tenns solely of the factors or attributes that enter into
an association, whereas quantitative analysis measures the amounts of
these factors or attributes and the degree to which they are associat-
ed with each other and with other factors or attributes called "effects",
and the rate at which these change in relation to each other - how
much of a variation of one is associated with a given amount of varia-
tion of the other. Mr. Cobb makes use of the.following illustration:(xx)

"In an economic study of the relation between
the interest rates on short term notes and the price
of longyterm bonds it was observed that bond prices
seemed Commonly to follow the note interest rate,
but only after an interval of time. This interval
was ascribed to the psychological fact that the
bond investor acts more slowly than the note buyer
and takes more time for consideration. This fact
was noted and the conclusion that bond prices
followed note interest rates was modified by the

 

G) Economic Journal of England, March 1928, p. '72.
(xx) Economic Journal of England, March 1928, p. 71.

13

P N'"

, l
statement.thux“there was an interval of delay«duo~tu' ‘“‘\
psychological "lag". In this case the psychological
lag was used aS“n,dntum'in the consideration of the
problem and was a qualitative datum because it was
not measurably or comparatively defined. Later, in-
tensive observations Were made as to the period of
time and a curve was put into the problem indicating
how soon after a change in note interest rates.the
bond market would move. The minute this curve Was
put down on paper the psychological lag ceased to
be a qualitative datum and became a quantitative
datum. The quantitative nature of a curve is not
dependent on its correctness or the conclusiveness
of the investigation which produced it. All that
is necessary to make it quantitative is that it 7
shall be so stated and defined that its relation ~
to the problem can be analysed and tested by some
standard and confizred or'corrected. The ascer-
tainment of the lag described is a measurement by
intensive observation of the psychologiCal average
action of many minds considering two different types
of transaction_ rem different standpoints. It is
nevertheless susceptible of conclusive determination
as indicated by the progress made by the economists.
The distinction betveen qualitative and quantitative
data is not a difference in the nature of the data.
it is a difference in the method of statement."

Some writers on method imply by their statements that quan4
titative analysis is more "scientific" than qualitative analysis._ A
better statement is that these condition and complement each other.
Quantitative analysis alone is virtuallv helpless, Qualitative an;
alysis alone is inadequate and unsafe in ordinary use. Considered
as a sole method, the qualitative has the important limitation that
it is difficult to state the terms cf it and the conclusions with
sufficient definiteness so that only one understanding results. The
result is a vast amount of dispute and disagreement occasioned by
nothing more than unconscious differences in definition of terms or
interpretation of the language of the generalizations. Another .
limitation is that important factors in the problem.may be overlooked;
and there is no easy way of matching the conclusions with experience
to show that they have been overlooked. Given qualitative analysis
of a high order, with rigorous definition of terms and statement and
testing of conclusions, and results of very great value follow. But
qualitative analysis of this high order is far from common. It is
practiced only by the best minds. Any one can try his hand at qual—
itative analysis. Nearly everybody does. People indulge in it most
freely in the fields in which they are least informed. Consequently
for the ordinary run of research workers, it is highly necessary that'
data be collected and qialitative conclusions checked at every turn.
Only after quantitative analysis is there any safety in the conclusions
with the ordinary research worker; and even the most gifted are oc—
casionally blind to inmortant elements in problems.

a

14

The difficulty of recording qualitatiwe data definitely and
'of reporting the steps in qualitative analysis, also makes it a less
~‘secure structure to which to add later. Checking and testing of con-
clusions is more difficult and uncertain.

The principal inadequacy of qualitative analysis arises from
its failure to weigh and evaluate the different elements in a problem.
‘Forecasting requires measuring the quantitative relation between causes
and effects.

There is no such clear—cut division between qualitative and
Quantitative as the foregoing would suggest. Much of what paeses for
qualitative is informally quantitative; that is, at least some rough
approximations are made of the significance of the different factors.
~ihere are minds so nearly qualitative that all factors seem to be
‘weighted nearly alike in their reasoning processes - but they are
indeed exceptional. The normal mind at least goes so far as to rank
the causal elements as to their relative importance. Such analysis
may not be numerical; but it is roughly quantitative. Even in the
matter of data, the distinction is not élear-cut. Are grades of
students qualitative or quantitative? Can shades of green be re-
duced tO'a quantitatve basis? Types of soil?

, An important question in this connection is as to the de-
sirability of division of labor between qualitative and quantitative
analysis. Shall we have certain persons of superior logical ability
concentrate on qualitative analysis to the exclusion of all quantita—
tive considerations; and let the more ordinary workers plod through
the routine of collecting and analyzing data? Or shall we insist
that none shall work without quantitative data? The first condition
is the one which comes nearest to prevailing at present; and constant
pleas are being made that we change to the second. Clearly there are
advantages and disadvantages in both arrangements. ”Most careful
students of the history of science will agree that it will be well

to continue to let a few of the best minds largely concentrate on
qualitative analysis. Specialization will be advantageous here as

in most fields of endeavor. No doubt, however, it would be well if
even these kept more closely in touch with quantitative developments
than some of them have done in the past. Moreover, for the usual
type of\research workers, no doubt the best program is one which
combines the two types of analysis.

Though much criticism has been leveled at the largely
qualitative analysts in the past few years, they are far from being
as deserving of it as the much larger group of workers who have used
quantitative methods without a proper qualitative grasp of their
'problems. It may be proper to have workers of this sort collect data
and go through the mechanics of analyzing it - laboratory research
has-its horde of common laborers too; but somewhere in the organization
of thb project, if it is going to contribute to progress, there must
be a directing mind which has thought it through in qualitative terms.
The greatest weakness of present research in agricultural economics is
that a sufficiently competent directing mind of this sort has too fre—
quently been lacking.

a.Indngtive'gg It is true that science is more obviously
' ductive inductive than it is deductive in its logic. The

.procedure is more conspicuously from a number-of

individual cases to a generalization. Professor

W. C. Mitchell, however, has called attention to
the blunder of identifying induction with the quantitative method. "In
all thinking", he says, "It is necessary to pass from the confused
data yielded by observation, and back pg particular facts. ‘(The under-
lining is ours). In this respect the method of R'cardo and the method
of the modern.quantitative worker are identical."%x) It would have
been even more important tO'haye pointed out that the first analysis
of‘a problem, that which isolates it and separates out its elements
pend defines terms and classifies, is much more largely deductive than
inductive. Its results are obtained by applying principles already
formulated to the problem in hand, by passing from premise to con- ‘
clusion by those rigorous processes which Pearson has described as
involving "logical necessity." The going "back to particular facts"
that Mitchell speaks of represents the testing of the generalizations.
This may be done as'a part of the piece of analysis in question, as'
when forecasting equations are applied to the individual cases and
the residuals examined; but it is more conspicuous in the continous
testing that generalizations receive from the moment that they are
made, by workers'over the whole field. '

The Scientific ' In broad outline, scientific method is the
Methods same in all fields. "The scientific method is

'one and the same in all of its branches," says
Pearson, " and that method is the method of
logically trained minds."(xX)

"The unity of all science consists alone in its
method, not in its material. The man who classifies
facts of any kind whatever, who sees their mutual
relation and describes their seQuences, is apply-
ing the scientific method and is a man of science."(xxx)

When these processes of classifying facts and discovering
their mutual relations and sequences are observed more closely, how-
ever, it will be found that they take different fonns, which are
commonly designated as different "methods."

1. Analogy.’ The simplest and most primitive of these is
the method of analogy, in which inference is made
from an example or two as to other cases or situa-

tions. A bad business depression accompanied a-Democratic'administrap

tion in the 90's; it would accompany another one in 1929.. The great
weakness of this method is that the cases are not likely to be suffic-
iently parallel.

 

(x) Proceedings of the Thirty—ninth Annual Meeting of the American
.Economic Association (1926) p.20

(xx) obt‘Citog P. 10.
(xxx) II II P0 12.

is

2. Case ' When the number of cases than Can be used
Method. as parallels increases a little, the pro-

cedure is likely to be called the case method.

In primitive use, the cases used have generally

been the striking cases. Much of our ordinary
thinking about business cycles even today is based on experience in a few
of the more climacteric of them. History of earlier times is largely in
terms of outstanding or abnormal events. In modern use, the case method
has come to mean a thorough internal study of a limited number of cases -
as one might study in detail the affairs of a few mail order_houses, or
farms, or communities. '

3. Informal Given still more cases, a mental procedure
. Statistical more nearly approaching the statistical is
.13 followed. Professor L. L. Bernard in an

article on "The Development of Methods in_
Sociology" has this to say of the "Informal statistical method":(x

"Informal statistical method involves the
rough generalization which is made by approxima—
tion from facts, possibly poorly defined and
attested, which are picked up in more or less
random observation. In this way the man in
the street makes up his philosophy of all types
of events from the weather to the presidential
election. He "guesses" at general tendencies
from what he sees or thinks he sees. The sample
he uses may not be representative and his method
of drawing conclusions may be anything but rigs
orously exact. However, it was the only method
of seeing many separate events together as a
unitary or single process, that is, of-getting
perspective and unity into one‘s view of social
life, before rigorous or mathematically exact
statistical method was invented. In spite of
the fact that it was the forerunner of math-
ematical statistical method and was universally
used for many centuries and still is used for
tentative generalizations, and by the masses
almost exclusively in their thinking, it is not
ordinarily regarded as a stistical method at all.
But such it is, and the only distinction between
informal and formal statistics is that the former
is merely more or less accurate approximation,
usually of a fairly simple sort, while the latter
is a process of complex calculation of general
principles and processes, of averages and means,
on an exact mathematical basis.

"Informal statistical generalization is older
than formal sicence, even then recorded culture.
It is the unconscious inductive source of many of

 

(a) G.U. Yule, "An Introduction to the Theory of Statistics", 1922
edition, p. 10. (Charles Griffin and 00., London).

17

the ancient maJor premises which the early phil-
osophers, such as Socrates and Aristotle, sought
to test by means of a more or less rigorous logic."

\‘4.; The Statistical Statistical method becomes formal when enough
“ Method. cases are included and measured in a definite

.. way to make possible the use of at least ele-
- ' mentary mathematical processes. These cases
. ' . need to be sufficiently alike so that they
possess'the same general order of attributes and common measures can be
applied to them. Statistical data. says Yule. "are quantitative data
affected to a marked degree by a multiplicity of causes." (I). The
problem of statistical method is to take data of this sort and determine
the relations between variations in them, and to make inferences front
these for a larger universe. The "Multiplicity of the causes" makes
it nece3sary to isolate the different causes and their effects.. Statis—
tical'method does this by taking actual phenomena, cases that have
actually occured;' It_has two_phasee. description and inference. It
may merely deal'with numbers of cases, in which case it is numerical;
or it may'attempt to deal with magnitudes, in which case it is
quantitative. . “

5. Egperimental . In the experimental method in its pure form.

' ' a special hypothetical set-up becomes the
basis of the conclusions, The effect of the
different factors is determined by holding
all conditions constant or uniform except the

one whose "effects" are to be measured. and a definite amount of change
in this condition is balanced against a definite amount of change in
the result. Sometimes it is merely the effect of the presence or
absence of a condition that is noted: in such case. the method is
qualitative eXperimental in place of quantitative. In practice. it

is often not possible to hold all conditions but the one constant, so
that the procedure followed is a combination with the statistical
method. Experimentalists are realizing increasingly that the two
methods often need to be combined.

In concluding the discussion of these five scientific methods,
it should again be pointed out that the logiCal processes are the same
for all of them. The experimental and the statistical method, for ex-
ample, both go through a process of isolating the effect of a given

variable. Although the actual procedures are Vastly different. the
logic of them is the Same. '

 

(x) The Monist. April. 1928, p. 310, Also see Black, Introduction
to Production Economics. D. 631. . '

18

,Historical ’ In some classifications of method, the his-
4 Research torical and the geographical are named as

separate methods. It should be clear from
the above that they represent a-classification
upon another basis than the foregoing.. The
'coordinate classification with historical is contemporaneous. The
logical difference between them is in the type of data used. Incident-
ally, because of differences in the type and sources of data‘ histor-
ical research makes use principally of the analogical, the case, and
the informal statistical method, whereas the statistical method is
used mostly with contemporaneous data, Professor Bernard would include
archaeological and anthropoligcal as also coordinate classifications
with historical and contemporaneous.

w
'Geoggaphical Geographical research clearly uses contempor—
Research aneous data for the most part — it is like the

anthropological in this respect. The special

feature of geographical research is that it

deals with data of local differences, in which
local differences figure to a marked extent, particularly as distinguished
from the data of time sequence. {L ‘

LA vast amount of data, however, is sufficiently independent of
- bothltime and place so that local and time differences can be largely
ignored in the analysis.

All of these methods and types of research will be taken up in
detail in the body of the handbook. The foregoing is intended merely
to serve as an outline of what is to follow. The'order which will be
followed is statistical, informal statistical, case, analogical, and
egperimental. Then will follow sections on the historical and
geographical as fields or types of research.

Social Science Although scientific method is the same in broad
Research outline in all fields, there are important‘
differences in details of procedure growing out
of differences in the data available or obtain—
able. Obviously the social sciences cannot
simply take over.ths methods used in the natural sciences. Neither,
on the other hand, can they develop methods that are entirely foreign
to them. In practice, it will seem that social scientists are giving
special attention to certain phases of procedure. But this merely
follows from the fact that these particular steps are more difficult in
'the social sciences. The attention being given to measurement of economic .
value by economists, or to mental measurement by psych010gists, illustrates
l this point amply.

The same statements apply when research in economics and the;
other social sciences is compared. In broad outline the method is the
same, but the details and emphasis at the various steps are significantly
different. For instance, the economist observes closely and attempts to

measure mass changes over relatively short periods; the historian ob-
serves them in a more general way over longer periods and usually has
to be satisfied with mere statements by contemporaries or roughly

' accurate measures. The economist devotes much time to examing the
representativeness of statistical data; the historian to the authen-
ticity of explict statements by participants in historical events or
contemporary observers of them. Research in agricultural economics
is confronted by the special difficulty that the units of observation.
the family farm, the local marketing agency, etc., are small and make
few records of their activities; also a large proportion of the econ-
,omic decisions pertaining to agriculture focus around the individual
farm as a center and are remote from actual markets where decisions
are translated into monetary valuations. The close relationship be-
tween the producing and consuming activities of the farm family is
a further difficulty.

Research method in the social sciences is still in the early
stages of its evolution, and although the problems which it has to
handle are essentially of the same logical types as those of the natural

, seiences, they are usually much more difficult at almost every stage of
the analysis, in the isolation‘of the problem, the securing of the
sample for study, the definition of the unit, the exclusion of extran— .
eous considerations, the reduction of the number of variables, the
measurement of the varying factors, the measurement of the results or

, the effects, etc. In fact, so difficult are many of these phases of _
most of the problems in the social sciences that it is an easy and
common assumption of workers in the natural sciences that social prob-
lems cannot be studied scientifically, that there is really no such
thing as social science. But it must be admitted that truth is
sometimes arrived at with respect to social problems as well as to
problems of the physical universe. Then there must be some method
by which truth is attained in such matters. It is the task of students
in the social sciences to discover and define and classify these methods,
_observe their essential modes of procedure, their successes and failures
and the reasons for them. Perhaps some of the conventional practices
of social scientists do not lead to sound conclusions. Perhaps many
things in the past have been affirmed as truths following researches
in this fie1d_which have later been found to be untruths or part
truths. But this is true of biological and physical science; it does
not mean that truths can never be successfully establiShed in the social
field. Instead it points to the need for a more careful study of
nethods. It was the realization of the foregoing that led to the
establishment of the Social Science Research Council. and later to the
'setting up of a special subsidiary committee to consider methodology
for problems in agricultural economics and rural sociology.

Economics Ks. While it is inadvisable that the lines between
Natural Science agricultural economic and other agricultural

,\ Research research should be sharply drawn, nevertheless
" a clearer understanding of the spheres of each
should help greatly in the administration of
research. -Research projects in marketing can belong either in the
field of econbmics or of applied natural science; but they should not

20

have the some content. The natural soienCe asyects of the marketing

0f flotat963. for example, include the scientificshases of disease con—
t01. standardization of product, sorting and grading, preparation for
shipment, care in transit, and storage. The economic sepects include
sources of supply. factors affecting sufiply, consumption, factors
affecting demand, factors affecting price, relation of price to pro-
duction and consumption. relation of price to grade and class. market
differentials, market organization, set-up of marketing business units.
°°St °f marketing and the like. All of these latter call for a ready
comfirehension of the principles of economics, that is, of the principles
relating to price and value and the organization of productita.eff0rt
for the maximum utilization of human and natural resources.

Even such duestions as grades have both their economic and their
statural science technique aspects. It is one thing t9 W°rk °ut.d°'
bC?iPtions.of market grades_that can be effectively used in separating
potatoes into grades and classes, and the necessary mechanical deViceS
for putting these specifiéations into effect. It is an entirely dif‘
ferent matter to analyze_consumers"preferences and prices in such a
way aSto determine what system of grades and classes. with what‘
dividing lines between them. will result in selling the crop f°r the
largest not price and satisfy the largest volume of consumers"want9-
The Bureau of Agricultural Economics has recently become aOPtely
aware of the fact that its past work in market grades has been too
‘ largely on a purely physical basis.

Suffice to say at this point that any organization of research
in an institution that results in having the soonomic aspects of prob-
lems handled by men trained merely in the applied natural sciences, or
the converse of this, is to be condemned. Detailed discuSSion of the
problems of organization of research arising out of the foregoing. Will
be taken up in a later section.

21

R E S E A R C H P R O C E D U R E .

mm. 1
Choice Of The first step, if the effort of the investiga-
' ' ‘ tor and the research funds placed at his dis-

Project (x) posal are to be most effectively applied, is
‘ - to choose the most desirable project. In-

' ‘ , vestigational organizations of many sorts have
been set up'in-response to the.general belief that a scientific attack
upon the economic problems of agriculture gives the only adequate hope
of their solution. [Such agencies are immediately confronted with the
problem of developing a plan of operation and-program of activity., _
To this work the utmost care shOuld be given, since time, funds, and
personnelhialmOst never admit of developing all parts of the field at
once, and even were this possible, the problem of utilizing the results ,\
of investigational work demands that certain important issues_be select-i.
ed for treatment, given adequate emphasis, and the results of research
brought effectively‘to'bear upon_a public which cannot be expected to
give attention to'more than one thing or a very limited number of
thingsvat a time. . . , '- '

Bases:’ The first task'thon for any research worker

215 Relative. in choosing an individual project. or in_
importance v _' shaping up the program of an organization of

which he has charge, is to weigh as fully as
. , possible the relative importance of the
several lines of investigation to which he conceivably might direct
. attention. Here he may well be guided by four considerations.

Firet, he may.seék to-determine the magnitude of the problem
or extent of interests involved. Within the state or other geographic
division to which his activities are related, a very much larger number
of people will probably be Concerned in one or a very few situations
or problems than in any other. Looked at from another angle, the
value of the financial interest at stake may be much larger in one
case than in another. The writer recalls an occasion in his own exw
perience when his organization was urged to make investigations in
connection with the problem of livestock marketing and that ofrxgycorn
marketing in one of the Corn Belt states. 'Very little Consideration
of the.relativo importance of the two.problems indicated that atten—
tion should first be directed to the industry which touched a major
source of income for a large part of the state rather than to the.‘
problem of limited and local scope.v In other cases naturally the
decision will be much less easy to arrive at, particularly if compli-
cated by other issues which may very appropriately enter into the
decision as to the choice of project. However, the'actual worh

 

(x) E G Nowill. Institute of Economics.

22

done both by institutions and individuals shows countless illustrations
of effort being expended on problems of comparatively small importance
while at the same time issues which involve large interests are
neglected.

2..TimelineSS' A Second consideration which bulks large in the selec-
tion of research projects is timeliness. While certain investigations
are designed Quite properly to add to our basic knowledge of economic
phenomena and have a comparatively slender tie to current happenings.
a large part of research work in the field of agricultural economics
and rural sociology grows out of maladjustments, uncertainties,fdr
unsatisfactory conditions which clamor for prompt diagnosis, and if
possible, remedial treatment,‘ While obviously the attention of re-,
search workers should not be directed to passing problems and ephemc
.ezal interests to the neglect of basic studies and forward-looking
programs of research, such agencies need also to avoid any tendency
toward other-worldiness and a neglect of current problems while the
satisfactiOn of purely economic inquiry fills their minds.

Timeliness frequently does not mean preparing for a program
of relief. The old adage, "strike while the iron is hot" should be
considered seriously, though sanely, by research workers in order that
they may make examination of important and acute phases of their prob.
lem, while they are in progress, when they can be analyzed to best
advantage. A singde illustration will perhaps suffice. If we were
to have really authentic information as to the character of the recent
so-called" land-boom" in certain sections of the United States, it was
important that research agencies should be on the alert to make first-

_hand investigation of a situation at certain distinct stages, particu—
larly as the movement reached its crest and broke, and thereafter as

, it appeared to reach the bottom of the subsequent decline, with values
stabilized or turning upwards. If we are to understand phenomena of
this sort, we must accumulate bodies of data’taken first-hand at the
moments when and in the places where such phenomena develop their

most significant phases.

Timelinesa may also be considered from the point of view of
general public interest of the fitting of the work of one investigator
or a particular institution into a larger program of work involving
other experiment stations, the United States Department of Agriculture,
or outside research agencies. It is hardly possible to place too
much stress upon the fitting of particular research projects so far as
practicable into general research programs if we are told to avoid the
danger of fragmentary, miscellaneous, and provincial studies.

3. Available A‘third major consideration in the choice of
resources. projects concerns the fitting of plans to

available resources. It is lamentably true

that in terms both of personnel, of funds,

and of collaboration with other agencies we
are all too limited and must fit ideal schemes to the practical limi-
tations_of particular times and p1aces.}

23

Paseing the obvious fact that researches must be scaled in
accordance with the funds available or which can be secured by inter-
esting parties particularly concerned in the results of the investigap
tion, a brief word may be said about personnel. The rapid expansion
of research work in agricultural economics and rural sociology over
a comparatively recent period has not permitted adequate training of
all those whom it has been necessary to appoint to positions on re-'
search staffs. Neither has it been pOSSible to carry out the process
of selection, weeding out, and replacement in such a way as to get
best results. It becomes peculiarly important, therefore, to devote
to tasks of planning, direction, and major strategy those persons
whose qualities, training, and experience fit them for this and to
utilize persons of less maturity and more limited training to routine
tasks or special types of work.

To this end, it is fortunate that so great a variety of
qualifications is called for in the gathering of field data, its
tabulation and statistical analysis, and the preparation of a variety
of written products designed to make investigational results available
in a considerable range of uses. Experience very soon shows that the
qualities which make an admirable field man, those which make a good
planner and director of a research project, those which make a good
statistical analyst, and so on, differ widely, and that too much
care cannot be taken in selecting and adjusting personnel to the task
or narrow range of tasks in which special qualities can be most effect-
ively used.

In this problem, of course, the difficulties of adjustments
are enormously greater where the staff is small, since, particularly
in state experiment stations, the range of problems pressing for
attention is always quite varied. Even in this situation, however,
much better results may be secured by careful planning, division of
labor, and specific training. But after all this is done, there
will still be the issue as to choice of those projects for attention
to whose execution the available staff can be most effectively fitted.
A small staff cannot afford to dissipate its energies over too wide
a field. It may have to reject some projects merely in order to
concentrate its attention properly.

To complicate an already difficult situation, stress should
be laid upon the choice of projects with reference to the availability
of co-operating agencies and the desirability of adapting any program
to the working out of a somewhat comprehensive plan in which several
agencies will collaborate at the same time, or at times which syn-
chronize sufficiently well so that the results will have their great-
est comparability and that a problem of wide geographic range may be
surveyed as a whole rather than piecemeal. Little more need be
said under this head than to call attention to the desirability of
co—ordinating research plans into such comprehensive programs, and
the fact that under the leadership of the Association of Land Grant
Colleges, the'Bureau of Agricultural Economics, the American Farm
Economic Association, and the Social Science Research Council, move—
ments of this sort are receiving more attention today than ever be-

24

‘ fore. ,Gobd teamwork is prerequisite to the attainment of the best
‘research results, and fitting the efforts of the individual or the
.particular'inStitution into such teanhplay should be given due weight
in the shaping of our research programs.

4. Q§g_3§‘product. A fourth approach to the general problem of
‘ ‘ selection of projects may be made‘from the
standpoint of use to which the product may
be put. Here we shall echo some of the
points made in discussing the question of
ti eliness above. A large amount of our research work is very_proper-
1y undertaken with a view to its immediate value as a guide to remedialv
activities. By and large. it may probably be safely admitted that this
A'}ill be the first consideration in the choice of research projects under
most circumstances. On the other hand, shaped merely to this proximate
end, they will lack in depth and permanent value. In many cases in-
vestigational activities shaped directly toward the thought of remedial'
measures will prove disappointing in results and perhaps entirely ab—
ortive owing to the fact that the general problem in which a oarticular
issue had its setting has not been sutfficiontly explored before the more
particular line of investigation was undertaken;

For this reason. particular attention must be given in any
continuing program of research work toward the laying out and system-
atic prosecution of studies of an oxplorative character. The precise
nature of issues to be given detailed investigation will frequently
not appear at all from a cursory examination of the field or not 5' .
be suggested by those who find themselves puzzled by or suffering frbm
certain difficulties of which they have intimate contact but very in—
complete perspective. For any soundly developing scheme of research
work, therefore, it is necessary that explorative studies be under-
taken not only at the beginning of the work but be carried along in
such a way as to check the significance and relationship of more
specialized studies and for planning the program of work for a suffic-
ient period in advance.

Finally, the complete development of research activities
must include the prosecution of searching studies into fields of
knowledge which do not promise to yield a product of immediately
discernible practical use, but which are designed to add to our
store of basic knowledge about the general phenomena of economic
and social life with which immediate problems are concerned.

Studies in statistical method as'such, with no clear idea of the

use to be made of results if secured, would furnish an illustration
of this type of research work, as would also both theoretical and
statistical exploration into the nature of the price—making process,
with ramifications into the field'of money and credit, commercial or-
ganization, and cyclical movements.. Psychological and even political
investigations of a highly theoretical character seem to be clearly
required if we are to have an understanding of the general situation

and forces within which our group activities develop.
. ‘V

25

If we are to build wisely and solidly for the future, there.-
fore, some of our resources and the time of some of our staffs who
show particular aptitude for the pursuit of truth for its own sake
must be allocated to studies whose practical value in any particu-
lar instance is problematic but to which as a. whole we must look
for~the development of the basic knowledge on which further progress
in the field must depend.

_‘ \25
PLANNING no: PROJECT

The next step after choosing the project is to
plan it. This will be tn en'up as a number of
smaller stops in chronological order. It will
be assumed that these steps all precede the final
preparation of the project statement, although
it is realized that in usual practice the project
statement is couched in such general terms, or is understood as sufficient-
ly subject to revision as to details, that most the steps here outlined,
are scarcely needed as preliminary to it. If such is the case, what
follows may be interpreted as further and more definite planning that
:should follow such project statements before any actual assembling 0T.
-enalysis of data is begun.

, Altogether too little of such planning is ordinarily done.
The tendency is simply to decide to study something, then get the data
together and proceed to analyse them. The results of this careless
and hasty planning are chiefly as follows:

1. Duplication of work already done elsewhere.

2. New results are not in form to be fitted into
the picture of work already done so as to
modify it as needs be and extend it.

3. Great waste results from attacking projects
that yield little in the way of definite
conclusions, even negative.

4. Much data is collected that is never used,
or is not in definite enough form for
analysis,

5. Data that are already available or can be ob-
tained in better form elsewhere are col-
lected from expensive and unsatisfactory
sources. This is a common fault of
surveys. The farmer is sought as a source
for everything.

6. Most important of all, data that are essential
to the analysis are not collected.

7. The method of assembling the data is not planned
so as to expedite subsequent analysis.

'~,‘ 8. The_senpling turns out to be bad; or the units
or measures prove to be faulty; or the
analytical attack breaks down: etc.

§;nl.,Examinetion of It is ascepted as one of the first essential
literature g5; procedures of sound scholarship in the natural
othcr exoerience_ sciences that before any new experimentation
in same 3321;. or analysis is taken, all prior work shall be

carefully examined. The report of the ex-
periment must contain a summary of this prior
> work in its very first pages. The same practice
is widely accepted in the social sciences also, although not so regularly
followed. The failure of agricultural economists to follow it carefully
is one of the reasons that fellow-departments in agricultural colleges
are often skeptical of their work. It is true that the problem is some—
what different in some types of agricultural.economic research. For
example, it is hardly possible in the opening pages of a report on a
survey to summarize all preceding surveys of the same or related types
in the world. .Nevertheless, before the project is started, all the
surveys in the same general area should be examined and their methods
of attack and results compared and differences accounted for if possible.
In projects more definitely of the research type; such as of factors
affecting prices, the natural science procedure can be followed exactly
in this matter.

Dr. E. W. Allen of the Office of Experiment Stations makes
an extremely pertinent criticism of many of the projects submitted to
him. He says that one of the major results that will come out of
many of them is that those working on them will get informed as to
the subject in hand out of facts and analysis already available.

Such is not research - it is learning the subject. But such learn—
ing of the subject is highly necessary as a preliminary to a real
project. The practical progrmn growing out of Dr. Allen‘s criticism
is that most of such preliminary learning should be done before the
project be submitted, so that the results of it can appear in the
definiteness of the project statement and the keeness of the analysis
of the problem it displays. The practical difficulty with such a
procedure is that if it is not finally decided that the project is
not Worth continuing, as should often be the case, there is no defin-
its Purnell project to which the time and expense of the preliminary
[work can be charged. Dr. Allen can no doubt suggest a practical
solution of this difficulty. One might be to carry a Purnell item
called "Preliminary work on projects". Although this might be

- abused, it would be better SO than to haye half-developed project
statements continually coming in.‘

It will aIWays be possible, however, for the senior members
of staffs, because of wide aCQuaintance with the field and methods,
to prepare project statements for the younger memters which will
satisfy all reasonable requirements in the matter of project state-
ments; and there will still be the need of having thosa directly
handling the project work up the field. In any event, semething
more thorough than is required for a project statement is needed
before the project itself is undertaken.

28‘

The Job of working up the literature and other experience in
the field should comprise the following: '

1. Discover what results have already been obtained.
2. Learn what methods have been used - units and measures,
sources, methods of collecting it, methods of ans

elysis, etc.

8. Assemble all data from other sources already available
that bear on the project.

If
g.
I

Learn the available descriptive and historical facts, _
that constitute the background of the problem and
show its relation to other projects and to the rest‘
of the'field of knowledge.

5. understand the deductive analysis of the problem as

worked out by others. » ._ l r
2. Preliminary The foregoing leads directly into the second
deductive stage in projectéplanning, namely a careful
analysis. rigorous deductive or qualitative analylis of

the research problem in hand. Set over against

this procedure is that of going out and getting
all the facts that may possibly beer on a general subject, in whatever
form seems most acceptable, and then seeing what can be made out of
them. It is argued by a few that this latter is the scientific pro-
cedure, that the other is likely to lead to certain conclusions en-
ticipated from the preliminary qualitutive analysis. The answer is
that all students of scientific methodology recognize that some sort
.of an hypothesis as necessary as a basis for all purposeful research;
that seldom does a scientist not have one even though he may be foolish
enough to deny it. The essence of the scientific method lies in the
accuracy and objectivity of the measurements and analysis, and not in
prowling around in the darkness feeling for mysterious objects. The
fact that some great discoveries have been accidentally does not prove
that only objectives of a broad genernl‘nature should be set up.

It is also a mistake to describe the deductive analysis which
is made of economic problems preliminary to quantitativesresearch‘upon ‘
them merely as the setting up of hypotheses. It is more properly de-
scribed as getting a clear view of all the factors and forces in the
problem, and thinking out possible relationships between them. There
has been at least two centuries of rather intense thinking as to re-
lationships in the economic world — and there is much of it still to
do. The applied sciences of ecbnomics offer a vast field for dis-
covering more of such relationships. As society becomes more complex,
new relationships come to the fore. It is possible for a person with
no experience in economic thinking, having had little or no contact

29

with the thinking in terms of economic relationships by the best minds
of the present and preceding generation, to envision many of the im-
portant economic relationships in a research problem; but he has to

be an unusual person. What a research worker poorly equipped in this
respect does instead is to single out a few superficial relationships,
sometimes already thoroughly discredited in the real economic world,
~and’couch-his-problem in terms of these. Even the man of the street
may be sounder on many points of economic analysis than those who
know just a little economics. For example, the answer which an or—
dinary housewife will make as to how much it Costs her to make her
childrens' clothes is likely to be better economics than that Of

some home economics students Who have been taught a way of figuring
it out in college. Likewise the ordinary farmer's statement as to
what the value of land is based on, than that of the economist who
has thought no further into the subject than merely that it is based
on the income from it.

Sound preliminary deductive analysis is as essential to
sound conclusions from economic research as is sound analysis of the
data. At the present stage of development of agricultural.econ~
omics, it is the more important. We have had tons and tons of data
collected without sufficient close thinking as to the relationships
involved in it. Even those accustomed to deductive analysis find
their research powers as often limited on the side of the qualitas
tive as on the side of quantitative analysis.

The research workers on a project should attack this part
of the preliminary work as follows: First, get all the help-on it
possible from other projects and from literature of the subject.
Second; follow back the economic analysis suggested into the general
literature of economics bearing on the subject. Third: Talk over
the problem with the best economists on the teaching staff in the
department of economics, if there is such a department. Otherwise,
seek help from the outside. Fourth: it will pay in many cases to
write out an analysis of the problem along the foregoing lines and
submit it to competent critics.

 

 

 

 

 

3. Statement of objectives. If the foregoing is
4. EstabliShment of boundaries— carefully done, the project
articulation with other fields. can be narrowed down so that
5. Defining relations-ips to other other steps, as follows, may
studies. _be quickly taken: First, the
6. Definition of the universe. objectives of it can be stated

in specific concrete terms -
in much more definite terms _
_ than found in the usual pro-
ject statement. To illustrate, in place of a statement in terms of
analyzing market business practices, one can be made which names the-
specific practices to be analyzed and the plan of attack on them. The
- farming system that is being analyzed can be definitely described to-
gether 1th the aspects of it that are to receive special attention.
Second: the articulation of the project with other fields of knowledge
can be stated definitely and procedure specified for cooperating with

.30

“yer securing the assistance of other departments whose field is related
‘to the project. For example, if the study is of factors determining
the price of butter in the local market, plans can be stated under
which the dairy product Specialists will analyze the butter as to
quality. Third: A clear statement can be prepared as to how this
project ties in with others already made or being made. and into the
whole subject of agricultural economics and the rest of the social
sciences. Fourth: The universe~which is to be studied can be defin—
itely described and defined. In the case of a.study of farming
systems, statements can be made as to what sorts of farms will be
excluded from the survey — e. g. - farm specializing in the sale

of pure-bred cattle in a dairy region. In the case of a price

study, the grades and classes of the commodity and the period of

time covered can be determined.

' l

-\

7. Units 229 The subject of units and'measures will be
measures. discussed in detail in Part II. At this

point, let it be said that the only safe

procedure is to write out a careful defini-

tion of all the statistical units used
that are special to the problem in any particular - what is to be
called a farm, a rented farm, a tenant, a cropper, a family, a house—
hold, etc. Measures need to be even more thoroughly analyzed and
described. Not a scrap of data other than for trial purposes, should
be taken until there is definite understanding as to what measure or
measures are going to be made in the analysis for size of farm, input
of labor, input of each of the kinds of feed and ration, for value
of land, interest rates, supplies obtained from the farm, volume of
business .of an elevator or creamery, intensity of production, etc.
There is no more difficult problem than that of measures in the whole
field of the social sciences.

8. Method of analysis. The leaders of the project should have their
plan of analytical attack so clearly in mind
that they can visualize the sort of tables
and graphs that will come out of it. This
does not mean that these tables and graphs ,

will ultimately appear in the report. The results may prove to be en»

tirely negative; or the data may yield only to some other sort of at—
tack. It may not be possible to foresee half of what is likely to
happen if the project is one which breaks new ground. Nevertheless the_
methods of analysis should be planned in advance as carefully as pos—
sible. The planning should include ways of testing the sampling, of
testing the units and measures used, of checking the accuracy of the
data, of isolating the effect of different variables, of deflating for
price level changes. etc.

81

 

9._.Preliminary exploration Before going further, it will
of the field 3: the be advisable frequently to cen-
universe. duct a sort of preliminary ex—

ploration of the field of the

universe which it is planned to

study. 'lhus,.if the project
involves studying leasing systems in detail in a certain‘area, it may
be well to know what all the leasing systems are, and where they are
localized, Before studying livestock shipping association problems
in an area, it will be well to know where they are to be found. and
the economic status of them in general. If this is already available
from earlier or other studies, this step may not be necessary. Often—
times, however, the other study is out-of-dnte and it is important to
have such general information as the foregping reasonably accurate as
to the current situation. The three main purposes of this prelimin-
ary exploration are as follows: (l).to test the soundness of the
decisions provided for above as to the universe, units and measures,
etc.. (2) to furnish a general background for the study and fill
in the gaps between areas or periods not included in the sample,

(3) to furnish a basis for the sampling or for choosing representap
tive areas;

The preliminary exploration can frequently be made from census
data and assessors' records and from official records of various kinds;
but sometimes recourse has to be had to questionnaires and other re-
connaisance methods of collecting data.

10. Sampling; We are now ready to narrow the
choice of area study down to a particular area
or period. or period of time in which we

are to make our detailed ex-
amination of the data. The
4 technique and philosophy of
this will be discussed under Research Method in Part II. There is
a very close relation between this and the interpretation of the re~
sults. It must be done with as full an understanding as possible
of all the implications. Sampling may not be pOSsible in many cases.
If the method of representative areas is followed instead, all the
consequences must be understood and appreciated.

The project is now in form so that the leaders can say it will
begin in a certain place at'a certain time, and proceed to prepare the
schedules and other details of execution of the actual project.

memos or mime/(xi

‘Or 5. ion ‘ 'finder‘thi‘s head pill be -» included/a few-observ/
' “ f,.,‘ ,.vations of\5"2§neral nature, Certain parts
‘ of it, such as the collection and analysis of
data. will be considered in detail under re-
search method. The choice of projects has
been considered in a preceding section of the handbook. 'There is a
close interrelation between the choice of project and the staff or-
ganization for carrying it out. It would indeed simplify matters if
for every new project it were possible to draw into the research or‘
ganization men well equipped with both theoretical and practical
{training and contacts for that‘particular'project. This would in-
‘ volve either a large research staff or that the field of research be
limited to a series of closely related projects.

\

In a smaller organization trying to cover a wide field of
work the research worker who changes from a project in grain marketing
to One in livestock marketing of neCSSsity has to spend much time in
making contacts and familiarizing himself with the surroundings and
Conditions of the livestock industry. If the change were to a prob—..
lem in taxation the Idjustment would be even more difficult to make:-
so difficult that it is doubtful if it should be undertaken.

In organizing the staff for work upon a particular project
two things are essential. first, familiarity with the field in which
the work is to be done, and second, a knowledge of the particular
method of study which is to be employed. Lacking either of these,
the conclusions are apt to be sterile if not misleading. If the two
characteristics can be found in the one worker so much the better.

If not, they should be secured by a combination of workers or ad-
visors.

With capable leadership and direction secured, the remainder
of the necessary staff may be recruited from less experienced in—
dividuals of aptitude. The double purpose will thereby be served of
economy, and providing training for less experienced workers. Aptitude
for the various phases should be censidered in assigning workers to
the various parts of the project. Some of the workers may be especial—
ly fitted for the field work, others for the compiling or interpreter
tion of data. An appraisal of the-capabilities of the various workers
is one of the first essentials of staff organization.

Budget. Past experience is the best guide in making
a budget. More often the budget is made too
small then too large. The problem is usually
that of making the means equal the needs. Un-
expected phases will turn up which should be pursued. A frequent error
is that of underestimating the amount of clerical help which will be
needed. Often a fairly good estimate is made of the amount of time re-

 

(xj J. I. Falconer. Ohio State University.

4‘.— ....

u“ e

33

‘ ‘ - ;

éuired for the field work but inadequate provision made for the compil-
ing and interpretation of the data. When a specific project is started
a~hudget should be made both for the current year and for the entire pro— ,l.—-
ject sosfar as the extent of the le.tter can be forcasted. Some degree-'9
of flexibility in the budget assigned to particular projects is always

- an advantage..

Establishing Contacts Before undertaking the study, contact
and Working Relations. should often be made with the leaders
5r . in the group from whom data are to be
secured and with those whom the results
of the study aré'especially hoped to

7 serve. Care in this regard will serve the double purpose of guarding

against pbssible future misapprehension as to the nature and reason,for
the study, and should yield many useful suggestions to the workers-in
shaping and developing the study as well as facilitating the extension
of the results when.completed1 In field studies, contacts should al-
ways first be made with the county agricultural agents of the region.
where work is to be done. -If work is to be done among organization
members, the leaders of the organizatiOn should be fully acquainted
with the plan. Establishing and maintaining these contacts is an
essential to any research group which hopes to fully serve the area

or group which it roachea.~ ' '-

Tentative,schedules.‘ ‘ f , .If the nature of the study requires a
schedule, the schedule finally used 1
will probably be the product of an ev—

olution. The first draft should be prepared by the leader of the pro-

Ject, after having made the necessary preliminary exploration of the -

subject and other research work done on it or in progress. He shOuld

consult specialists in related departments about the various details
of it and frame his questions according to the suggestions. He should
have the final schedule criticized by his Colleagues who have had ex—
perience with schedules and by subject-matter specialists in related-
fields.. Then he is ready to have it tried out in a preliminary way;

A few schedules should be filled out and a preliminary analysis and

interpretation made. This will frequently uncover gaps in the data,

points which were sought but canno+ be secured, questions'which are
frequently misunderstood by the one interviewed; or the preliminary
study will reveal additional data, much desired but not known to be
available when the study was originally planned, or that the schedule
is covering too wide a range of subjects. If several workers are

'using the same schedule, another product of such a preliminary attack

will be that each worker will have been taught to give the same ins
terpret ation to the various questions of the schedule. Many a field
study has been completed only to find that the various workers.eere
giving different interpretations to the same question. . -
Before revising the schedule, the data secured in the pre-
fiiminary study should be examined as to its adequacy to serve the
purpose needed.vat may be found.that-adequate_data are not obtain-
able on many points in the schedules, or that the answers have-been
recorded in a way that prevents their being tabulated or interpretated.

...

,.-- ‘\ _ __ \__

"“ “\u‘. -1,» .-._..,__....

34

If the project is a large one, it will pay to take enough schedules
in a preliminary survey so that the results can be tested by fre—
quency distribution analysis and the other tests of sampling. Other
POSSible tests may be such as the following:_checking expenditures
against receipts, checking feed consumption of livestock by feeding
standards, checking individual schedules against assessors‘ reports
when available. A conference of all the workers in which each ques—
tion is analyzed thoroughly will also help.

Collection 9: data. There may be one or many engaged in the
collection of data. If only one there
is the merit that the data will more
likely be collected and entered on a
comparable basis, and that one person

will be more intimately acquainted with all phases of the data. If

mahy are Collecting the data, however, the work will progress much
faster. In the latter case some one person should be in charge of
the group and check each schedule as it is completed. In this way
errors can be uncovered and differences of interpretation between

the various workers detected. When this plan is followed, workers

of less experience can be employed in charge of an experienced

leader. Expense is thus reduced and training given to new workers.

The number of workers which.can with advantage be employed will de-

pend upon the nature of the schedule. If the work is largely routine,

a large number may be employed; if however, it is a matter of ap—

praisal, better results will be secured with fewer workers or more
experience.

Recognition should be given the fact that frequently the
worker who will excel in the interpretation of data may not excel in
its collection. This may make desirable a division of labor. Un-
tactful work in the collecting of data has diminished the usefuleness
of many a study and many a research organization.

Appraisal and When all the data that are available or
planning 9: analysis. at that time will permit have been as—

 

sembled, an appraisal should be made Of

its adequacy. This should engage the

attention of both those who collected.
the data and of those who- are to analyze them. A thorough appraisal
of the data before working them over will save much time and wasted
effort in their analysis and interpretation. While frequently the
error is made of attempting to get more out of the data than their
extent or character would justify, a more frequent error is in not
fully utilizing that which is available. _Such an apiraisal may dis—
close gaps which- still exist in the data or may reveal possibilities
of analysis beyond that which was originally intended. If possible
to do so. these gaps in the data should be filled, so that the final
results may be more complete. Or as will be more frequently the case,
such an appraisal will reveal fields for further work.

FollOwing this appraisal, the plan of analysis should be
revised as necessary.

35

._Drawing conclusions. All through the analysis, but especially
toward the end, when the results begin to
appear. interpretation is called for. This
requires vision, imagination and caution

as well as a knowledge of the field of work and the methods of collect-

ing and analyzing the data. The judicious weighing and summarizing of
the results is the capstone of a research project._ The tragedy of
many a project has lain in the collecting of a large volume of facts
which were not later adequately analyzed and carefully interpretated.

The subject of inferences from data will be handled in a later section.

Wherever possible, first inferences should be checked by further an-

alysis of the data; as well as by submitting them to other workers in

the same or allied fields. No inference which appears unreasonable
should be accepted merely on the basis of the data.

PART ONE
W
A STATISTICAL METHOD.

Statistical method is taken up first partly because it
“gig the main resort of agricultural economists and partly because it
~isimnch"simpler to present the other four methods after then before
statistical method.

be) What is Statistical Method.

Statistical method includes the collection, summarizing.
analysis judgment and presentation of numerical data of-mass pheno~.
menon. x The first part of this statement names the five main'
divisions of statistical method. There is a question as to whether
this statement includes as specifically as it should the matter of
choice of units and measures cnd measuring. The data of statistics
are numerical. although this feature is not conspicuous when the
method is applied informally with a limited number of cases. The
phenomenon observed usually involves the behavior attributes of a
large group of individuals. Hence the "mase phenomenon." This
statement is most clearly applicable to timeaseries changes. Seldom
does any economic change take place that does not have wide ramificap
tions involving whole masses of the social substance. Statistical‘,
method is also involved, however, in such problems as summarising
phySical measurements or the results of simple psychological exp
periments. Mass as applied to such situations means simply a
number of cases. ,

As already pointed out, the essence of such data is that t
"in the actual universe, nothing is same, nothing can ever be act- }
ually repeated." They are like this because in each individual
case there is at least a slightly different combination of attrib-
utes or causes. No two sunrises are exactly alike, to take an
extreme case. But thus far none of them has been sufficiently
different to change the main feature of the event. In social
phenomenon, these differences are always affecting the main event
being observed, sometimes even preventing it altogether.

, This characteristic of statistical data furnishes star
tistics with its essential problem, that of "generalizing im-
perfectly comparable data." The method of statistics for handling

 

(x) Professor Warren M. Person's statement.

\

‘- 37

this problem is to give to these "unlike or related items a common~
_denominator and thus a common classification or.perspective."(x)
The farms in an-area surveyed differ in a thousand particulars.r
Statistical method abstracts certain attributes common to all these
farms, and analyzes variations in these and associations between
them. The different marketing years for potatoes have many special
attributes. Statistical method abstracts certain of these. The
problem of statistics-is to make some statements on the basis of
these abstractions that in the first place describeythe sample, and
in the second place indiCate probabilities for a largfir universe.
The difficulty of this latter is that the rest of the universe is
never exactly like the part of it that has been examined. "'

The following excerpt from "The Grammar of Science" shows
the relation of statistics to logic:

"Such a contingency table as we have
schemed above is the numerical syllegism of ob—
servational science, which replaces for all its
purposes the barren syllogism of the old Aristo‘
telian logic. We do not say, 'Some of B is A,’
but we state numerically how much of each class
of B is associated with each category of A. In
actual practise, of course, it is impossible to
form a table of the whole population or the whole
universe of A and B things. we take here as
elsewhere a "samplb" to illustrate that universe,
and we have to take great precautions not only

' that this is a true sampde, but that our inferences
from the sample may be applied to the universe
under discussion. The theory of samples-—their
probable errors and legitimate use~-is the chief
topic of modern scientific statistics."(§i)

There is, however, this difference between the syllogisms
of logic and those of statistics ~ those of logic, once the premises
.are accepted, lead to incontrovertil&d_conclusions, whereas those
of statistics, as already pointed out, predicate only probabilities.

.’ '

Statistical‘pgbcedure assumes premises also. 'These enter
in partly by the way of mathematics and partly in connection with
the nature of the phenomena. Statistical method uses mathematics
as a tool. There are mathematicians who go so far as to identifir'
statistics absolutely with mathematics, to say that statistical
analysis leads by incontrovertible steps to the same iron-clad con-
clusions as formal mathematics. The economic statisticians, how—
ever, generally take the position that the mathematics of sampling '
and error and inference thus far developed, which holds rigoroulsy
only for pure chance and "simple samples" of entirely unrelated

 

(1) Bernard, 0p. cit., p. 314, 519
(xx) "The Grammar of Science", p. 160-161.

58 '

‘events. is inadequate for the‘needs of economic phenomena. and that
there is little prospect of mathematical analysis soon being developed
which will be adequate. Once the assumptions of puregphance are '
violated, inference has to proceed along other lines than those based
on simple mathematical probability. The nature and extent of the
relatedness between the events, or occurrences of attributes, become
significant factors in the generalization for the sample itself and
the inferences from the sample for the larger universe.‘ Probable
inference gives place to statistical inference;(x) Mathematics may
in time set up formulae comparable to those of probable error and
probable inference for certain significant typical sets of events
other than those of pure chance; but it has a long task ahead if it
expects to provide for all the contingencies which social phenbmona

, offer. When mathematics does so provide, it usually sets up so many

unreal assumptions that the formulas are of little applicability.

To the extent that statistics makes use of the mathematical
formulae of pure chance, it must accept all their assumptions. How
the legic of mathematics starts with asSumptions and introduces more
at nearly every step, is not generally realized by the ordinary man-
ipulator of formulae. This is why men trained to a limited extent
in mathematics with no understanding of the philosophy of it rely so
implicitly upon formulae when they enter the field of statistics.

(b) Units and Measures — Definition 9: Terms.

The importance of having definite and clearly stated sta»
tistical units and measures and terms and concepts has already been
sufficiently emphasized. The purpose of this Section is to state
the principal considerations involved in setting these up and to

.apply these considerations to the field of agricultural economics.
Problems of this kind are especially difficult in the social sciences
because of the great variability of individuals and the complexity
of phenomena. This is amply illustrated by the efforts of public
agencies to define income and determine the value of public utilities:
of economists to agree upon such concepts as land, capital and profita.

_ Developing such units and concepts is one of the major tasks
of any young scienCe such as agricultural scenomics. The workers in
new fields are always eager to set up units and measures and concepts
and to have these standardized and generally accepted. No argument
is needed as to the importance of having a large number of these in
common use and generally understood. The sooner such a condition
is reached in any science the better. But there are also great
losses from too early standardization. Progress along this line
must come as a process of evolution, each new generation of units
and measures and concepts being different from the last, but out of
this process coming slowly the stating of certain ones in terms that
change but little thereafter. It is unreasonable to expect that the

 

(x) See article by E. B. Wilson in Science. Mch. l926. Also note by
same Journal of American Statistical Asseciatien,,3une 1927.

39

a

‘first products_of such effort will long prove to be entirely satis—
'factbry. ->If they should happen_to come into wide use at the start,
the time of‘theirfbeing discarded or modified is likely to be post-
poned too long; and progress seriously impeded for a while before

and Just after this happens.' For reasons that need not be dis—
cussed, such early standardization of certain concepts and measures
occurred in‘the field of agricultural economics, with the.usual con-
sequences. There is always danger also that the need for comparable
results over a wide area or a long period will unduly’influence de-
cisions in this respect. Although uniform practice may begnecessary
in a large grOUp of similar projects in many states, this muuttnqt
interfere with setting up improved units and measures for new groups
of projects later. ' - '

For the present, therefore, the great need in agricultural
economics in this respect is a careful analysis of units and measures_
and concepts whenever a new project is outlined. especially one which
-is likely to be repeated in many states,and choosing the ones which
are best qualified to serve the desired ends; and along with this an
extremely careful definition of all such units and measures and cone
cepts, so-that others can follow the results closely. .Emphasis upon
this latter rather than standardization is the need for the present;

.Qnii§_2£_gggg- Let us consider first the type of statistical

42§2£12£ unit that is involved when an enumeration or

count is made, as when the census is taken,

or automobile accidents are recorded. With

our increasing efforts to record current
developments, this is becoming an important feature of research. What
shall be defined as a "sale" when transfers of farm real estate are _
recorded, or as a "forced sale"? Shall large tracts of cut-over land
be classed as farm land in such sales records, likewise lake-shore
property? Census enumerations furnish the outstanding example of ths"
need of such units - a farm, a farm family, a mortgaged farm, a tenant,

etc. Following are a few important considerations with respect to
such units: .

1. All possible doubtful or border—line cases must be an-
ticipated before the enumerating and recording is begun, and a defini-
tion, and frequently a set of instructions in addition, prepared for
the guidance of the enumerators or recorders, and for permanent record.
This usually requires that the definitions be tried out rather ex-
tensively before they are put into final form. For example, such
cases as the foregoing may need to be provided for in the definition
of a farm family: a son who lives with his parents and operates an-
.other farm nearby; a family that lives in town but has a farm in the
country upon which nearly all the labor is hired, the husband having
more or less of a city occupation besides. If the latter is ruled
out because the family is "not living on the farm", what will be
done with those families which live in a village but drive out'to
the farm to work? The reports of such enumerations should include

' a full statement of the procedure followed.

2. Such definitions will often need to be arbitrary to a cer—
tain degree. For example, a family which earns its living partly from-
factory work and partly from farming Operations, living possibly either
in a village or on a farm nearby, has as much right to be classed as
having manufacturing as farming as its occupation. In an enumeration,
it must obviously be classified as one or the other, although sometimes
double counting or an intermediate classification is possible. Arbitrary
demarcations which are established in such cases should be logically
consistent with the purposes of the enumeration, follow current usage,
and be workable. These several requirements may, however conflict with
each other. For example, current usage may be illogical or inconsis-
tent with the purposes of the enumeration. As an eanple of inconsis-
tency is the exclusion of rented farms from the mortgaged group; bUt
thus far no other arrangement has proved workable.

3. The units should be named according to current usage if p05—
sible, so that when the data pass into popular use without the techni~
cal definitions attached they will not be used incorrectly. There is
a Question as to whether the federal census' definition of a farm does
not include a different sample of establishments than the popular con-
cept. particularly inclining more to those with inconsequential farming
Operations- The classification "gainfully employed"’is commondy recogr _
nized as a-misnomer. "Income" is a dangerous term to use in any except ‘
the currently accepted sense — as witness the extent to which ”labor
income" data have been used by the general public as evidence of agri—
culture's well-being.

4. In planning units and measures, some attention should be
given to the probable use of the data, the education and understanding
of those who will use them, and their disposition to biased interpre—
tations. If per—capita farm incomes are likely to be compared directly
with per—capita urban incomes, the definition of "gainfully employed
in agriculture" should take this into account if possible.

5. Comparability is much more important with units of enumer~
ation than with other statistical units and with measures. The use
of census data and official records is largely in comparisons between
areas and periods, and if the units are different, the data may be
nearly valueless.

Complete comparability is of course impossible in most cases.
A good definition of a dairy cow in one period does not fit as well in
some later one. The larger the area included, the poorer the fit of
the unit in any part of it. The world census of agriculture will raise
many difficult questions of this nature.

41

 

“gaits of ' , The foregoing discusSion assumes a
Samples - definition complete count of all individuals or
of the universe. cases. Much of statistical analysis

“I

is based on a limited number of cases

taken at random or selected from the

. whole number. .Thisgintroduces a few special considerations. In

'the first place, the nuniverse" set up to be sampled is often less
,broad than that enumerated. Studies based on samples be representa»
tive selections are likely to be confined to a certain class;'or to
exclude the sorts of cases that are unusual in the area. Eor:examp16,
_it may be limited to dairy farms only, or may exclude certified milk
farmS. This makes the need for definition of'the'unit especially
.urgent. Such studies can be made to.conform more nearly to scientific
standards; less attention need be given to popular consumption. Com—‘

parability with other data is much less significant, but still import—
-ant in its place.

Definition of the universe is one of the first things to be
done in planning an investigation.‘ If the collection of data is to
be a part of the study, the universe should be defined before the
collection of data is begun. If secondary data are to be used in
the study, the universe should be defined before the data to be an-
alyzed are selected ' J fl'“

a To be sure, problems may arise in connection with the collec-
tion of the data that will necessitate a revision of the definition of
the universe: and when secondary data are used, the lack of data drawn
from the univerSe as originally defined mey_make a revision necessary.
For example, in a study of incomes from farming and their relation to
various factors undertaken by the Bureau of Agricultural.Economics
some years ago, the original intention was that the universe should
include all farms in the United States, and that the data would be
collected by sending questionnaires through the mail to the crop re-
porters of the Bureau, asking them for the details of their business
'for the preceding year. An analysis of the replies to the question-‘
naire showed, however, that practically no reports had been received
.from tenant farmers, due to the fact that very few tenant farmers are
casked to serve as voluntary crop reporters. It was further found
that very few replies were received from farmers operating small-
farms. Since it was impracticable to obtain comprehensive mailing
lists of tenant farmers and of farmers operating small farms. it was
.necessary to revise the definition of the universe and include in it
lonly ownerebperated farms of more than ten acres in in size.

~I . This study might have-been begun, as many have been, without

..'haring attempted to define the universe to which the results were to

'\

apply, but if so, and if the data secured had not been studiea to
determine if all classes of farms were represented, the investigation_
= Would never have known to what extent he was warranted in generalizing.
, from the results of his sample.

-... M- .. ._- ”74- us- a.
_ . ,». .

42

To paraphrase a passage from Mr. Bowley's text on the
“Elements'of Statistics: One of the first things to be done is to
determine what is £9 33 measured, and one of the last things to be
done.is to determine whatwhas been measured.

Subgfiiasses Definition of units is obuiously closely
'“ » associated with the problem of classificar
tion. In the case of an enumeration, the
sub-divisions of a main class must be so
definedcfihat: (a) They are mutually ex-
clusive, and'(b) include Somewhere all cases comprised under the
definition of the main class.

The problem of sub—classes arises frequently in connection
.qith types of land or soil or grades of commodities. The differences
‘Between classes may be essentially quantitative, but lack of a measure
compels resort to qualitative distinctions. The problem of defini-
t‘ion ‘then takes on them of market grading.

Mggfigggg. Statistical analysis usually involves the
further steps of measuring some of the at-
tributes of the individuals falling within
the classes defined, grouping these attrib-
utes into frequencies, and in many cases

discovering to what extent and at what rate variations in amounts of

wthese attributes are asSOciated with variations in amounts of §ther
attributes. Perhaps supply and price are to be measured and asso-

_ciated, or rainfall and yield. Some of the least—solved of all

problems in economics lie in this sphere,(x) At least two problems

are always involved in such an undertaking, one, the‘definition of
the attributes, and the other the finding of a unit of measure for

it. Thus efficiency has to be defined as an attribute before a

unit of measure can be found for it. In many cases, the attribute

itself cannot be measured, but some approximate indicator of it is
taken instead ~ as death-rates are sometimes accepted as a measure

of healthfulness. This is the third problem connected with measures.

Let us first consider the problem of definition of the at-
tributes. Teke size, for example. Is it to be a physical concept
or an economic concept? If a physical concept, is it to relate to
mere area or volume, or to power to turn out a physical product, or
to ability to utilize a physical input? The answer must depend
upon the objective of the analysis. If it is merely description,
then the more frequency distributions according to different con—
cepts of size, the more complete the description. If, however,
certain relationships are to be analyzed, then the concept must be
defined in consistency therewith. If, for example, size is wanted
as a measure of capacity to use other factors to advantage, the
definition will probably need to be in terms of the input factors.
If change in.size of the farm unit from decade to decade is desired,
perhaps acres or value of output will serve best. -

 

(x See Professor Irving Fisher's article on the measurement of?
utility in the recent John Bates Clams Memorial collection.

.43

, The concept of efficiency is one which is loosely used.

Strictly defined, it refers to the output returned per unit of input.
In"efficiently managed farm household is one which yields a large
output in satisfactions per unit of the cash expenditures, farm
supplies and household labor that goes into it.. Productivity is a
mere comprehensive term, taking account of capacity, or the amount
of the input factors that are handled, as well as of the input per
unit of output. ‘ v ; -i. .

Cost is a term that usually needs defining. Applied to a
production problem, is it the value of a certain fixed inputythat
proves to be necessary to call forth production - that is "necessary'
expense", or is it merely the value of whatever inputs happen to
occur?. Applied to living, is it the value of a certain "necessary"
budget,_or is it the value of goods actually consumed in any given
case? The term cost is used loosely in both senses, often in the
same study. ' " '

Value in the economic sense is the most difficult of all, ,
attributes to handle, and at the same time the attribute which economic
statistics is called upon most often to handle. As with many other
attributes, it frequently is not possible to measure value with suf- ‘
ficient precision for scientific purposes. Either some other why'must*~'
be found for handling the problem, or it cannot be solved at all.

Public utility commissions and other such public agencies are frequently
constrained by law to make vaulations, and the courts have under such
pressure developed a rough'set-sf working rules and definitions to be
used in such situations.. But no economist pretends that the values

so obtained have even approximate scientific validity in the maJority
of cases. The rulingfimmfticalphilopophy of the courts has more to do
with the valuations than any scientific considerations. It is beyond
the scope of this handbook to discuss theories of value. Obviously

all that has ever been said that illuminates the subject is pertinent
here.

Another attribute which needs defining is "the standard’of
living'I of a family or group, now most acceptably defined in the
words of Professor Davenport as.the goods which we feel the deprivap
tion of if We do not have them - a sort of customary plane and content
of living. . '

The foregoing illustrate the problem of defining the at—
tribute to be measured. Next is the problem of the measure for it.
The efficiency of a producing unit may show itself in output per unit
. of imput; but the efficiency of the input factors affects the size of
the product too, and there seems to be no wayttelling how much of the
output is due to the efficiency of each. Practice has been to at-
tribute all to the factor whose efficiency was being analyzed at the
time - to attribute it all to the land at one time, to the farmer at
another, to the cows another, and to the good varieties of crops at
another. Another procedure has been to assign values to theother,~
inputs and use the residual as a measure of the efficiency of the
'one in question. Thus "labor income" was conceived. It could be
used with equal validity to get farm efficiencgiand cow efficiency'

J.lf"asoatisfactory‘velnntion could be obtained- for the proprietor‘s
“contrihutdon to the product. '

'1
“Cost of living" defined as the sum of the values-of goods of“

actually consumed, has been used both as a measure of efficiency 0f~

rural living and of the plane of living. A low "cost of living" may,
indeed reflect efficient household management. But'it cannot at the‘
same time indicate a low plane of living. '

According to the most generally accepted analysis at present.
the value of goods entering into production is to be measured in terms
of the other alternative uses to which they could equally_well be put.

, These other uses could be working at other employments in some cases,
producing other crops in other cases, selling in the market—place in
ad here, leisure and recreation in others, etc. It is highly important
that this principle of value be applied in a realistic, that is, scien—
tific wage In the case of cash expenditures for productive goods,
there is no question about the reality of the alternative use value.

It can be assumed that the market price reflects a balancing of all

the alternative employments. The other employments, however, are not
so real in the case of proprietor's labor, If the problem under cone
sideration is necessary price for farm products to secure an adequate

'volume of production, the comparison of alternatives centers around
the choice of each particular person as to whether he wants to be a

v farmer or not. The alternative use value in such a case is what each
farmer could earn in other occupations, or perhaps in other systems
of farming. In this comparison, however, differences in living con—
ditions, prices of food, farm—produced supplies, preferences as to
occupation and a score of other things mus t be taken into account.

Seven hundred dollars on the farm may outveigh twice that in the
city. If the problem under consideration is not whether or not to
keep these men farming, but rather the more usual one of the best

‘ way 33 utilize their effort 23 their present farms, then the only
realistic analysis is in terms of other employments while remaining
on these farms, and these employments in most cases are in the pro-
duction of other farm products.

Likewise, family labor assuming the second type of problemp
utilization while remaining on these-farms - will find most of its
other employments at other lines of-activity on the same farms, al-.
though occasionally there will be an'otherwise equally desired alter-

native of working for a neighbor or in a factory or mine nearby.
Valuing family labor at what it would cost to hire an equal amount
of work done is a misapplication of the principle of alternative-
use value”; It does not represent the value of the alternative
employments of these particular workers. ,Marketable farm—produced
feeds are properly valued at what they will sell for in the local
markets if, and only if, first, a system of farming in which these
particular feeds were sold for cash would maintain itself over a
period - otherwise the requirement "equally well" is not met; or,
second, a system would maintain itself in which it-was the practice
to sell for cash When the market was right and not at other times.
Any other scheme of valuation is hypothetical and unreal.

45’ -..“

. ”‘Thesaixernativeause'Valueamrwanmaa~rarm-1and in any gfiVen“
use is the net value of its product..differences in effect on plant
food supply, in effect—on—soil texture and weed and disease control,
etc., having been included in the net value. if planted to the most
available other crop for the same place in a crop rotation; 0T per~
haps as based on the same analysis applied to whole crop rotations
taken as alternatives.

There are problems of a similar nature in using the alter-
native-use value as a measure of the value of the labor of chores
and field work.-of labor at different seasons of the year, and even
at different days of the week - in case it is desired to value the
inputs of different farm products separately. The alternative—use

values on a farm may even vary during the day according to the
weather. ‘

- Alternativenuses are certainly different for each farm.
It is more necessary to adjust valuations of agricultural productive
agents according to conditions on each farm than it is valuations
of factory productive agents according to each city. The individual
farm is the immediate market center for valuations of the productiVe
agents used on a farm - except when the problem is that of necessary
price to keep up the supply of farmers;

The third problem is that of using indicators of the at-
tributes in cases where the attribute itself cannot be measured.
Thus rate at which land is cleared may be used as an indicator of
progress of individual settlers. Obviously it will not describe:
the progress of those farmers who give more attention to buildings
than to land clearing. Number of families having heating systems
or bath—tubs in their homes-may be equally poor measures for the
plane of living of different communities. It is not well to rely
upon any one measure of this kind; and sometimes it is possible to
use a considerable list of them.

The essential requirements of good measures are that they
shall relate closely to the attribute in question, and that varia-
tions in them shall be strictly proportional to variations in the
attribute in question. Suppose the problem is to compare the
climate of areas as to suitability for farming. Average frost—
free period is one measure commonly used. But it has several
defects - one that it counts too heavily the cool but frost—free
springs and autumns of lake—shore areas; another that it does not
adequately include effective heat and sunlight. Perhaps the follow—
ing simple measure would be more adequate in many cases: number of
years in twenty in which corn matures before frost. '

Composite Efforts are constantly being made to set up
Units composite units — such as animal units, pro-

ductive work units, adult equivalents, total

cost per unit. There are great dangers to

be avoided in such composites. One is that
of confusing or burying the attributes or conditions which it is de-
sired to study - as a total cost per unit may largely confuse the.

/
1'46
”1 “variations in feed rations, which areuusually the most essential thing
*V,L\ .to analyze." tAnother is that/the basis of combination may not relate
;.~ ” T to the attribute being studies — for example,-the adult—equivalent-
‘-sfi. unit used in rural living analysis may be based on food consumption
.[ bhly, A physical basis of combination may not fit in economic an-
\ alyefsw and vice versa.

é 9222312: The measures used often take the form of co-
3 isnigr efficients, that is, ratios - such as the

birth rates, death rates, suicide rates and

the like with which we are familiar. Research
,' workers are constantly setting up new ones
rsuited to the needs of their projects. Even the conventional ones are
.-constnntly revealing weaknesses in use; and the new inventions often
prove entirely untrustworthy. Statisticians have developed a number
‘fig rules that assist as safeguards if observed closely. Every co-
efficient in effect has a numerator and a_denominator. The numerator
in a death rate is the number of persons dying in a given period, and
-the denominator is properly a certain number of persons, taken as a
base, who were exposed to the event of death during the period. The
rules mostly related to the choice of the-numerator and denominator.

, 1. Select the quantities used as numerator and denominator,

»in each case, so that the qutients derived may be legitimately com-
pared.(i) ,To illustrate the intensity of dairying in different regions
might be measured in terms of coefficients'heying either number of
dairy cows, or milk production in gallons. or butterfat production in
pounds, as numerator; and having either land in farms, improved land
in farms or crop acres as denominator. No one of these, as a matter
of fact, would furnish a legitimate base for a comparison. Percentage
of farm income from dairy products would give different results than
any of the above, but would not take account of the smell total pro-
duct of rough or sandy areas. When no coefficient can be devised _
that meets requirements, it is best to use two or three different 32%;,
ones. In other cases, a reasonably close approximation can be
obtained by "refining" the group included in the denominator until
it represents conditions closely identical with those specified in
the numerator. Thus per-capita incomes of rural and urban groups
become more comparable when reduced to equivalent age and sex
distributions. or geographic conditions.
Per-capita farm incomes for the nation as a whole reflect If“ *
southern conditions much more largely than do per—capita urban in- ’ {»
comes for the nation taken as a whole. The geographic conditions
are not equivalent.

The are or period included in the denominator should be con—
? sistent with the subject of the numerator. For example. to compute
_ percentage of land in a certain crop and include Considerable land
i where the crop could not possibly be grown would vitiate any Com—

 

?t; (E) King, Elements of Statistical Method, p. 42.

\
\

\
\

47

parisons. A rate of taking land into farms might properly occlude
forest.reserves or even 1arge_sections of swamp, desert or rock

‘ outcrop.

. The scope of the numerator and the denominator should be
as nearly identical as possible with the attribute or relationship
being examined. Thus if intensity of cultivation is the attribute,
the data and measures'used should reflect this only.

Coefficients which mix several attributes or relationships
should be avoided if possible. For example, "Crop acres per man"
may be high because of growing laboruextensive crops of doing is;
latively little crop work per acre on a given crop (e.g.-cultivating
the corn only a few times) or of effecient use of man labor (large
machinery units, keeping men employed, etc.) Some of those are
essentially opposing relationships. Gerrelations between crop

.acres per man and output are therefore likely to give inconsistent
; results.

a The names applied to these coefficients should be carefully
chosen. For example, the "purchasing power" coefficients heye often
been misunderstood. '

The foregoing does not introduce discussion of the standardh
ized coefficients which have been developed for general statistical
measurement - arithmetic means, modes, index numbers, standard divia~

, tion, coefficients of variability, correlation and regression, etc.
. These will be handled in their proper setting in the handbook.

(0) Errors in'Data.

Statistical literature contains altogether too little dis- '
cussion of the problem of error in the data themselves, of the types
of such error and their causes, of methods of testing it, and of the
effect of it upon the validity of conclusions. The usual textbook a
in statistics distinguishes between biased and compensating errors
and lets the matter rest except for an occasional Warning that
"accuracy in the original data is assumed." ‘ ‘

\ .

The distinction must be borne in_mind at this point between
primary statistics, those collected by those making the study, or
under their direction, and specifically for the purpose of their study;
and secondary statistics, those Collected by other agencies and for
other or more general purposes. Census data, market quOtations,and
data of carlot shipments are secondary statistics. \

» Let us first consider errors in primary data and list some
,of the important types of it. The whole subject will be taken up
more specifically later when the various methods of collecting data
are introduced.

\

48

l. Misunderstanding of the question or definition - See
discussion of this under ”Surveys” and "QueStionnaires". Data af-
fected in this way must in most cases be thrown out. "

2. Closely related to the foregoing, reporting for a
different unit or a different set of facts than those intended.
Thus an inquiry about taxes may be understood to mean property
taxes by some, all taxes by others.

3. Overlooking part of what should be included in the
answer. Inquiries relating to such things as receipts and expendi—
tures, number of livestock, etc., are most likely to be affected in
this way.‘ No doubt the census onumerations generally omit import-
ant items of production and sales. The usual protection against
this is to list all the items entering into the result separately.

«"

4. Confusing diflerent years - reporting 1927 taxes in
place of 1926. This becomes 5 rious if data are'wanted for a
series of years.

5. Failure to remember amounts correctly - a bias upward
or dDanard. For example, crop reporters may generally overstate
last year's crops as compared with this year‘s.

6. Intentional under - or over—statement, because of fear
of possible relation to taxes or effect on market prices.

7. Unconcious psychological bias - tendency to exaggerate
that which is in the center of attention. For exanple, the county
agents and bankers in a cut—ever region where a land-clearing campaig
was under way reported almost as much land cleared in two years as the
census showed for five.

8. Simple errors of estimate, such as of crop condition, or
distances to town, or weight of animals - the kinds which usually com-
pensate.

9. Round number estimates - often compensating in a large
sample.

10. Eistakes of actual measurement or counts.

Errors in'existing secondar1 data may have arisen in any of
the foregoing ways, and in addition the following:

1. Recording of transactions or operations.

2. Reporting of observations — illustrated best in the case
of price quoting.

3. Computations of totals, averages, etc.

4. Transcription from one record to another.

119.

~- One who uses secondary data must study thenrcqreruaavuens¢1s~
covez_all possible errors in the obtaining and recording of the data in
the -first place, and in the later work upon the data. .

(d) m (x) "

”The theory of samples" says Karl Pearson, " - their probable “

errors andllegit ate use - is the chief topic of modern scientific
statistics."‘ 31 It is such because most of the data we work with

pare only a sample out of the whole "universe" in which we are interested.

We cannot take time to analyze the activities of all the dairy .farms in
the world, or even in a single region. Accordingly we study d smal1
number of them. How true of the whole group is what we have- in this_
small sample of it? This is indeed a major problem of statistics.vv
The case is even clearer with time—series data. All we can possibly
have'in this case is a sample out of the infinite.

In some cross-section studies using the data of the census
enumerations..it is possible to include the whole universe, Such
operations are statistical even though no sampling is involved.

Sampling error, like errors in data, is a subject inads~
quately discussed in the-textbooks in statistics. ‘ The early American
textbooks. like King and Secrist, gave only passing reference to it.
One. has to go to Yule's Introduction to the Theory of Statistics to
find a discussion of it which is at all sufficient for careful scien.
tific work in the field of agricultural economics; and even this dis-
'CuSSion leayes large gaps in analysis still needing to be filled. (xxx)

(”The principal difficulty with all these discussions is that they talk

in terms of perfect or "random" samples only, and most of the samples
which we have to deal with do not fulfill these requirements. .Ws
need to know how valid are the conclusions under various practical
conditions of sampling. Unitl measures are set up more applicable
than those hypothecated in the assumptions of random sampling and
pure chance, we shall have to rely a great deal on empirical tests.

As a matter of fact, the term sampling is almost too narrow

_ to include all the practices followed in the matter of obtaining a

small number of cases to use in lieu of the whole. Folleudtg is a
list of the practices followed in cross-section studies of the
economic phases of agriculture:

1. Pure random sampling — as when the names or numbers
designating all the farmers in the region are written on slips of

”paper and a given number of slips is drawn. (from the hat.) 9

 

(x) op. cit. p. 161.
(xx) Dorothea Kittredge,EUniversity of Minnesota, reviewed'this
this section for the Committee. .
(xxx) The following recent textbooks have useful discussion of the
problem of sampling: ,
Jerome, pp. 13-23 165—178. .
Bowley, pp. 277-286 (Elements of Statistics).
Crum and Patton, pp. 118-125,196-210.
Mills, pp. 548-561.

5

5O

2. Every third; fifth, tenth or some other ordinal of the
farms is checked off on a complete list of all farms in the region
constituting the universe, and only these are visited. ,'

8. “Typical" areas in the region are selected and all farms
in the areaimarked'off are visited except the ones rules out as outside
the universe because abnormal.

4. "Typical" areas are selected and the field men work out
from a central point till they think they have enough cases.

5. Like the above, except that a good many farms are omitted
because hard to reach or small, or hays other activities mixed-with
V farming; or because the proprietor would not give the data.

‘ii 6. Like the above, except that the area selected is sectioned
or banded and only thGSe sections or hands epe'covered.
‘\ 7. The whole region is laid off in bands or sections of un-
iform size and all fanns in part of these bands or sections are coveaed.
at What the Bureau of Crop and Livestock Estimates call a
.“stratified” sample is taken - that is, the area is laid off in die-v
tricts and a certain number is taken from each district.‘ The dis-
tricts may be counties or townships. The number taken may be pro-
portional to the importance of the district, or the results may be-
weighted on that basis in case the number of returns is not properly
proportioned.

9. The stratification may be not in terms of geographic
areas, but of the various sets of conditions found in the area. Thus
each of the conditions of nationality, type of farming, and soil in
a state might be represented in a sample of rural living in a state.
The representation may or may not be preportional to the importance
of these conditions. '

To. What seem like "typical" farms, or marketing business
units, or rural villages, are selected.

11. An attempt is made to select a "typical" array of farms
or marketing business units - some of all the important types and of
all degrees of success.

12. Bankers or other local people are aSked to send in lists
of "representative" farmers, landlords, etc.

13. The cases select themselves according to whether or not
they respond to a questionnaire or keep the records desired or make the
desired peribdic reports. ~

It is impossible to say categorically that one of these methods
is better than another. The answer depends upon the objective‘of the

51 ‘ — -‘."_"

inquiry. Some of the objectives for cross-section analysis..c1ass~"
ifiedaaccording to pertinency on this point, are as follows: ._—-‘~

1. To get an accurage description of conditions, as of
.rural living, farmer's incomes or costs, or crops bein.g grown. With
such an objective, Nos. 1,2,7 and 8 offer the best basis for ade~
quate samples. although some qualifying statements will be made later
on these methods Obviously Nos.5 and 6 and 10 to 13 will not serve
at all; Nos. 3 and 4 will serve poorly; and No. 9 only a little
better. ' ~ ‘

é 2. To obtain a measure ofc mgg_ in conditions rather

t an absolute conditions. The absolute level obtained in a sample
may be too high. but the change from year to year may conceivably be
normal for the whole group.

3. To obtain evidence of‘relationships between attributes
or variables and to measure the extent of this relationship and the
regreSsion of one on another.

4. Merely to list the conditions that exist in a region:2
for example, the leasing systems to be found in it, or the different
varieties of corn. being grown. One may be interested merely in know-
ing that certain plant diseases occur in an orchard without knowing
exactly the relative prevalence of each. The question may properly
be raised as to whether such data are statistical. I

The practices in time series "sampling", if the term can 1
be applied to such data, can be briefly stated as follows: When
yearly averages are used. ordinarily all the years are taken for
which data are available. Sometimes, hoWever, a'cutgoff as of a
given date is used in place 01‘ the yearly average. yearly, monthly
and weekly figures may be either averaaos, or cut—offs on a given
day or series of days. Daily measures may be "closing" price or
some other cutuoff. “Average cost to packers", is an example of.
a daily average. The quality of randomness is obviously lacking
whenever cut-offs are chosen; but this does not necessarily mean
look of representativeneSS. All 01‘ the four obieotives for croSs-
section analysis apply to time series analysis, but not in exactly
the some way.

Si r“ It seems advisable for the sake of the later
Semelingh ' discussion to restate here in as simple a form
as possible the requirements of a simple _§gp_g
as given by Yule. (x) It should be stated first of
Illthe four objectives above named - namely an
accurate report of conditions in the universe being studied. A sample
adequate for this will meet the requirements of the other three object-
ives also; but it may be possible to meet some of the other three with
less than a "simple" sample

 

(1) Ch. XIII "Simple Sampling of Attributes", Yule prefers the tenn
simple to random because more than more randomness is required.

l The general definition of a random sample is one "found in'

such a way that every one of the individuals in the larger group has
the same chance of being selected in the sample, and that the selec-
tion of a particular individual does not influence the chance of
selecting some'other individual". (x) '

Yulefs three requirements of a simple samgle specifically
stated are as follows: ‘ w

(1) :"We assume that we are'drawing from precisely the same
record througheut the experiment." (p. 336). ‘

If conditions Vary between lbcalities included in the universe
being studies, the samples of it must all be so taken as to include all
these localities, Applied to periods of time, there can be no differa
rgce in the period included in the samples if conditions change in any
”essential way daring the whole period being studiod.‘

The effect of violating the requirement is stated by Yule as

follows: , '
f If we do not draw from the same record all

the'time, but first draw a series of samples from
one record, then another series from another record-
with a somewhat different mean and standard—devia»
tion, and so on, or if we draw the successivef _
samples from essentially different parts of the “'
same record, the standard error will be greatly
increased.

That is, "the standard-deviation observed will be greater
than the standard—deviation of simple sampling.": (p. 288).

How much it will be increased cannot be stated — it depends
upon the amount and the relevancy of the other variations. Yule states
that it "may be increased indefinitely as compared with the value it‘
would have in case of simple sampling." (p. 348), The type of series
which violates this requirement is known in the literature of sampling
as a Lexis series. '

2. :"We aesume not only that we are drawing from the same
record throughOut, but that each g: our cards at each-drawing may be
regarded quite strictly as drawn from the same record (or from ident-
ically similar records)." (p. 336).

"Consequently, if our formulae are to apply in the practical
case of sampling, the conditions that regulate the appenrnsce of the
character observed must not only be the same for every sample} but also
for every indiyidual in every sample. This is again a very marked
limitation. To revert to the case of death-rates, the formulae
would not apply to the numbers of persons dying in a series of samples
of 1000 persons, even if these samples were all of the same age and
sex composition, and living under the same sanitary conditions, unless,fly

a“

a. ""

(x) Drum and Patton, p. 205.

\ 53
further, each Sample only contained persons of one sex and one age.
'For if each sample included persons of both sexes and different ages,
the condition would be broken, the chance of death during a given
period not being the same for the two sexes, nor for the young and
the old. The groups would not be homogeneous in the sense re—
quired by the conditions from_which our formulae have been de~
duced." (p. 260).

- Yule states the effect of a departure from these conditions
of simple sampling as follows: .

"If the ayerage chances are the same for each universe from
which a sample is drawn, but vary from individual to individual or
from one sub-class to another within the universe, the standard-devia—
tion observed will be lees than the standard-deviation of simple
sampling as Calculated from the mean. values of the chances."

(Po 288). -

From this Yule derives ihs statement which applies to
stratified sampling (pp. 34e-o):

"If we are drawing from the same record throughout, but
always draw the first card from one part of that record, the secbnd
card from another part, and so on, and these parts diI‘fer more or less,
the standard error of the mean will be decreased. If, to vary our
previous illustration, we had measured the statures of men in each of
n different districts, and then proceeded to form a set of samples
by taking one man from each district for the first sample, one man
from each district for the second sample, and so on, the standard—
deviation of the means of the samples so Iormed would be appreciably
less than the standard error of simple sampling.

The result is perhaps of some practical intores.t. It
shows that, if we are actually taking samples from a large area,
different districts of which exhibit markedly different means for
the variable under consideration1 and are limited to a sample of
n observations; if we break up the whole area into n sub-districts,
each as homogeneous as possible, and take a contribution to the
sample from each, we will obtain a more stable mean by this orderly
procedure than will be given, for the same number of observations,
by any process of selecting the districts from which samples shall
be taken by chance. There may, however, be a greater risk of
biased error. The conclusions seem in accord with common-sense."

The type of series which violates this requirement is
commonly referred to as a Poisson series.

3. "we assume that the drawing of each card is entirely
independent of that of any other." (p. 336).

Yule's illustration of this is as.follows (p. 261):

"Reverting to the illustration of a death—rate, our formulae
would not apply even if the sample populations were composed of persons

54

of one age and one sex, if we were dealing, for example, with deaths
from an infectious or contagious diseaser For if one person in a
certain sample has contracted the disease in question, he has in—
creased the possibility of others doing so. and hence of dying from
the disease. The same thing holds good for certain classes of
deaths from accidents, e;g. railway accidents due to derailment,

and explosions in mines: if such an accident is fatal to one person
it is probably fatal to others also. and consequently the annual
returns show large and more or less erratic variatidnsr"'Cx) \

A The effect of positive correlation between the drawings
is to increase the standard error, that is, make the observedW:.~v
standard error greater than that of simple sampling; the effect'*; _
of negative correlation the opposite. The application~of this i
' to time series data is obvious. Ordinarily the events of one '
period are closely correlated with those before or after.
'Independence is almost entirely absent.

If all of these three conditions have been met in-selecting
the sample, the_series is usually referred to as a Berneullian series,
In series of this type. it is possible to apply the formulae which ‘
have been derived for samples, and find the limits within which sta«
tistical measures of different samples in the same universe will
fluctuate. Mills summarizes the result by saying: "This means that
'we may apply to the population at large statistical measures secured
‘ from the study of a sample not with confidence in their perfect ster-
bility, but with fairly.definite knowledg of the margin of error"”

involved in thus extending our results."€:;) Unfortunately; however,
we cannot'be certain that the conditions of simple samplingthave been
fulfilled in an actual case. In.fact, in the field of agricultural
economic data, the circumstances in connection with the'selection of
samples of much of the data make it plainly evident that the con—
ditions of simple sampling could not hays been met. It is there-
fore extremely important that the research worker in this field be
alert to these situations, in order that he may avoid the error of
temploying with confidence the mathematical formulae for probable
errors (applicable where the conditions of simple sampling have

been met) to situations where these conditions have not been'met;

If'a succession of samples is taken from a universe, the
means from the samples will ordinarily form a frequency curve of the
"normal" type, 'The mean of this frequency curve of means will very
closely appreximate the actual mean of the whole universe. It will .
do this even though the phenomena themselves do not yield an essential-
ly normal curve. The foregoing is what furnishes the basis.for using

 

(x) Yule distinguishes between these three requirements of simple
sampling by saying that the first means that the dice shall
always be thrown in the same way, and the second, that the
dice must be all exactly alike. The third requirement would
be violated if some of the dice were left down after each ‘
throw. See Secrist, pp. 400—406, 1925 edition.

(xx) Mills. page 553. 3'

I ,

55

Standard Error and Probable Error of the Mean as a measure of sampling
error. A given figure for Probable Error of the Mean is a statement
that the mean of the sample taken stands a fifty-fifty chance of being
this much more than or‘lesS'than the mean of the universe from which

it is taken. Increasing the Probable Error means getting a sample with
a mean that is farther sway from the true mean. ‘

Throughout this discussion on sampling. it has frequently been
necessary to refer to the "observed standard deviation", and a question
may arise as to whether this means the standard deviation computed for
the one'sample taken, or whether it refers to the standard deviation
observed from successive samples. -Theoretically, it refers to the latter
condition; practically, to the former. A simple derivation of the fore
mula for the standard error of the mean will serve best to demonstrate
what this means. ‘

Let us assume n samples to have been drawn from a universe” and
the standard deviation of each to be represented‘EVOT Q5 0‘)....o (RR-t Q
The standard deviation of the sums of the samples wil béog_:“7g+CK%+43uflk.
The standard deviations of each sample will tend to be the same in the
long run and identical with the standard deviation in an indefinitely
large sample, drawn under the same co ditions.* thgrefare, since all
tend to be the same, we write: =_ ,o~- = ‘ is.is the
standard deviation of thSaZums, andzghgostagdard dg$?gxiggwhf the mean of the
.of these n samples, will be l/n-th of this amount, Orqfilz h which
may be simplified by the following stamp/77“: :Sbl-D Ora/K: ,
Strictly, the standard deviation appearing in t e numerator of th ”for—
mula is the standard deviation for the fundamental distribution, but as
that value is unknown, the value of-the standard deviation observed for
the single sample is taken in its stead, on the ground that it approx-
imates the standard deviation of an indefinitely large sample.**'With
the formula thus transformed for practiCal purposes from the use of a
standard deviation for the fundamental distribution to that of the
single sample which is being analyzed, it also becomes necessary to
replace the \f??, the number of samples, by the \I7V_,the number of
cases in the single sample. IThe formula actually used for the standard.
error of the mean.is(fia=.$:3 ,where 0‘ is the standard deviation of- '.
the distribution in the ongfigample?**The probable error of of the mean
is .6745 times the standard error.

Obviously the less the dispersion, measured by Standard Deviation
(square root of sum of squares of deviations from the mean), the less
the Standard Error. That is, the more bunched the distribution around
themean, the less the Standard Error. The Standard Error also decreases
according to the square root of the number in the sample.

 

* Yule, p. 844. '
** Crum and Patton, p. 209; Yule, p. 353.
*** Mills, p. 555. .

’56

There is clearly no evidence in these measures as to the
universe from which the sample is taken except that of the Sample it—
self. The evidence is all internal. On two assumptions, however,
first, that there is considerable uniformity in the world, and second,
that the sample is representative and adequate, it can be reasoned
that the frequency curve and the mean of the sample should fairly ' "as.“
closely approximate the frequency curve and mean of the universe it—
self. The definite figures for the extent to which they do this ‘
which are derived from the formulae, are however hypothecated in the ‘\
three further conditions of simple sampling laid down by Yule. The
requirement that the sample be not too small (Yule p. 353), must not
be overlooked at this point either. What figures indicate the extent
to which the mean of the sample describes the mean of the universe
from which the sample is taken when Yule's three conditions are not
fulfilled, or the sample is too small, or when it is derived accord-

_ th to each of the 13 procedures in sampling outlined above, we can— I
‘not state at present with any degree of precision in most cases. '1
Yule himself says: "It is evident that these (three) conditions very
much limit the field of practical uses of an economic and sociological
nature to which (the) formulae can apply without considerable modifi-
cation." (p. 262). Neither do these measures take care of the errors
in the observations themselves.

_ The application of sampling principles to the various methods
of sampling will be taken up in more detail under each of the methods
of collecting data. But a few general observations are needed at
this point. First,many of the studies of farms, especially those of
the route type, and of marketing business units, do not include enough
cases to make frequency curves. statistical averages and other similar
summary expressions of much scientific value. The methods of analysis
to be used are those of informal statistics and the case method, to be
discussed later.

The same statement must in general be made when the individual
units for study are chosen as "typical“, or so as to constitute a typic-
al range or array of conditions (Hos. 10 and 11).

There is, however, a definite relation between sampling method
and the universe to be studied. It is entirely logicei and scientific
to plan to study only certain types of farms or families or marketing
units in an area, and to select those types carefully. The summary\_
statements obtained in such a case, however, must be clearly made to
apply only to those types designated for study, not to the area as
such. Besides, there is always the possibility that theltypes ommitt—
ed may be needed to explain some of the circumstances of the types
selected for study. The small or off—the-road farms ommited from a
survey may be needed to eXplain some of the conditions prevalent in the
area as a whole.

When areas are selected as "typical", the sample may be
adequate for these areas, but the areas themselves are not likely to
represent the whole region properly. The Probable Error of the Mean
may be a fitting measure of the immediate universe of the survey, but
not for the whole region. .

‘ ‘ ‘ "\"Nm
, _.,.— M

The requirement that the sample be drawn so as not‘tcgbe
taken from certain parts of the universe is a Lifficult:.one to meet 9*w5
in geographical samples in which certain areas are chosen for study, s
As more farms are included in one area, new conditions may be reachmdz— . ‘~s
poorer soil, or smaller farms, or a new foreign stock.

This difficulty and others pointed out above makes it highly
necessary to define the universe very carefully in advance of the
study.

If Objective No. l, describing conditions as they are in the
whole universe, is in view, sampling methods No. 1, No. 2, No. 7 or ,
No. 8 would almost seem necessary in any study where geographical con-
ditions vary much. Any of these may fail to meet the conditions of
simple sampling, although No. 7 is more likely to fail than Nos. 1 and
2, and No. 8 definitely fails in part. No. 9, stratifying the sample-
according to conditions, may meet the requirements of sampling in con-
siderable part; but the dangers of bias with this method are very
great.

All methods of obtaining cases which result in certain types
being more frequently included than others - especially methods Nos. 10,
ll, 12 and 13, and to a lesser degree No. 4 and 5 ~ give results which
do not describe the universe. Moreover, it is not usually possible to

determine with sufficient accuracy the extent to which the sampling is
b1 3866-.

If, however, the objective is No. 4, it is entirely possible
that such methods of sampling as Nos: 10 to 13 may serve. They will
even serve for many specific purposes included under Objective No. 3 -
that is - there are many sorts of relationships than can be studied
by appropriate methods even if the sample is not exactly representap
tive.

No doubt, also, a high degree of representativeness is not
so much needed for Objective No. 2, measuring changes.

58

- (e) §gurces of Qata and Methodgpgf Sgcuring it.

(1) The Survey-Method.

The subject of the use of the survey method has been handled
in the following manner. First, Professor Misner of Cornell sent out
an inquiry to all the experiment stations to discover present practices
in survey-taking. This inquiry is summarized in the pages following;
Professor Misner has also presented a statement of what seems to him
the beet survey procedure in farm management research. ’ Since its
first survey of Tompkins County in 1908, Cornell has taken a total of
93?3 farm business survey records, over a tenth of the total number
in United States, and has published 32 bulletins based on such records.

’ Second, Mr. Hawthorne of the Bureau of Agricultural Economics
has made certain analyses of data of recent surveys, particularly of
two in Ohio and North Carolina in which he participated. He has also
assisted the Committee by giving it access to his records and analysis
of farm business surveys as reported in the.1925 Yearbook (Table 652)
and kept up-to-date in his files.

Third, frequency analyses were made of the_answers to certain
types of survey questions to see what evidence this would furnish as to
the types of question which do and do not secure usable results.

Present'Use of ProfessoI‘Mlsner.hchcmxnnarized the present
Survey Method. use of the survey method as follows: Of

123 farm management projects reported to 52

the Social Science Research Council in

1927, 51 percent were classed as using
the survey method, 40 percent as using route methods, and 9 percent 4
as using secondary data. (Survey method was defined as "the obtaining
of information by personal visits, with the use of a definite schedule
The Collection of information from voluntary or supervised records when
practically no information was obtained in addition to that contained
in the records, was not classed as surveyxmethod"). Of the 80 market-
ing projects, 73 percent were using survey methods, 3 percent route
methods, and the remainder secondary data or data of private or public
records. Of the funds available for farm management research, 45
percent was used in survey studies, 48 percent in route studies, and
7 percent in the others. Of the funds available for marketing re-
search, the comparable figures were 73, 2 and 25 per cent respectively.

Of 50 survey studies reported as being planned by 44 experi-
ment stations, 27 were in the field of farm management, 10 in marketing,
5 in rural sociology, and.8 in land economics, taxation, credit and
other miscellaneous fields.

59

ggport upon Preliminary work — The following phases of
the inquiry. such work were reported:

1. Review of related studies and literature on the special
problems involved.

2. Analyzing all existing data.
‘ » \

3. Discovering\supplementary sources of data.

4. Stating objectives and special problems clearly.
5. Analyzing these‘proolems qualitatively.
‘ 6. Planning types of analyses to be made.

‘ 7. Listing information needed.

,\ ‘ 8. Preparing schedule.and testing it out in the field.
9. Deciding upon sampling method.

'10nder of questions on schedule: Ehe conventional farm
business survey order as outlined in most of the replies was the
following?

1. Name and address first.

2: Order in whigh the cooperator visualizes his business.
3. :Not necessarily in the order of enumeration.

4. Logical sequence with reference to schedule as‘a whole.
5. Questions about related infennationvglaéed tOgether.

6. Supplementary information at end of schedule.

The following other suggestions were made by workers with ex-
perience with other types of surveys:

1. Put simple questions first.

2. Place first the general and less personal questions, and
those not dealing with money.

8. Questions which stimulate interest should come first.

4. Questions likely to be resented should come last.

60

Arraggement of material on blan.:

1. Arrange so that the blank will be as convenient as
possible for the field work as well as for the checks
ing and sumari Zing."

E. Have a number of central headings designated by numbers
with the items in tabular form and separated into
blocks by heavy ruling.

Hd-sh-h

Supplementary sets of instructibns:

where were 21 replies to this question, of which 9 said
that oral instructions were given to the field men and 12
said that written instructions of varying nature were gdven
field men in addition to the oral drill work that they commonly
considered necessary to prepare inexperienced men fer collect-
ing data in the field. Some replied that while no written
instrudtions had been given heretofore, it was a mistake and
that in the future it was proposed to provide each enumerator
with written instructions. -

'7
Jr

Questionsghot best to include:

'1. Those involving much calculation on the part of the
I -- cooperator.

/ :2. Those about his financial status, mortgages, indebted-
. ness, charities, personal habits and other questions
which are of ultra—personal nature.

, ‘8. Questions to which definite answers can not be made,
for which no reasonable answer exists, or which in—
volve estimates for which the informant has no basis
to guide him in makin ng the answer.

rSome expressed the opinion that a tactful field man can get
answers to most personal questions as long as they are sensible
and_capable of being answered.

Both office and field forms..

The 34 replies tabulated indicate that 21 have printed field
and office forms and 6 field forms only, 7 not always both. The
veins of many of the survey records increases with age so that
it is desirable to have. in some instances, printed office
sheets on heavy paper to which the data may be transcribed.and
kept as a permanent record that may be used many years later.

”If’ ‘~ .

61

.,»""' Contacts with coooerators:

Some make soliciting trips qnd arrange for a visit with the
cooperators at meetings. ornir: pay the cooperators a surprise
visit. ~53ne institutions send out introductory letters prior
to visits. Sons use tke telephone or other means of making
definite appointments With oooperators. The desirability of
these methods will depend upon the circumstances.

Size of field crggg:

The answers were to the-effect that one person Should
supervise the work in the field and check the records as'
they are brought in, when enough-enumerators are employed

Ito require or justifiy this. If only two or three men are
in the field, they may check of the work of each other.
Ordinarily_two men to an auto was considered a most efficent
crew for field work. (Cornell recommendS'three or four).

-- Preparation of crew for work:

1. Have them prepare and work up summaries'of similar
,data.‘ ' ' i ‘

2. Have them study publications, tabulations and completed
records of similar work.

3. Drill them on method of approach in asking questions.

4. Have them listen to an experienced person taking a record
in‘tho field and work with him in completing and work~
.ing'up same. , , ‘, ~ .i

5. Hare them take a record in the field and receive the

criticisms of others who listen, including an experi-
enced person who is supervising the work.

Checking. in the field:

,It was generally considered that records should be checked
for missing information and errors each day, in a central
office by the leader who is to be responsible for the prepara~
tion of.the results for publication. A memorandum should be
attached to each record indicating errors and omiséions so that
the enumrator may have the same next day. He will then quickly r
discover the questions he is getting wrong or omitting and may
correct any error or supply any missing information when passing ‘
the cooperator in his travels the next day. Promptness in checks g
ing recdrds is very desirable as a way of increasing the accuracy.

62

Sampling:

The answers t0 this question show that little real considers» t“.
tion is given to it at most stations. A few of the significant answers
were the following:

One man attempts to get records for one-fifth of the farms in
the county in sections midway between the best and poorest. An-
other prepares a list of cooperators and takes every fifth men r
the list. Another simply aims at 200 or more records. .In con-
trast to this, one takes 20 to 30 records in each of mamyartefin
One person endeavors to obtain enough schedules so that in 6 to
10 intervals the data will show normal frequency distribution.
In some instances comparisons are made with census data, as-

. sessors‘ information, previous surveys or any other existing

.‘F data that will help in indicating whether the sample is fairly
representative of the area to be studies. In most cases an
attempt is made to avoid actually "selecting" the cocperators.
It is considered desirable fit most institutions that hand—picked
cooperators be avoided, as these are likely to comprise the

best only,and not give a fair sample.

Selecting areas:

The answers to this question were even more unsatisfactory. an
Most simply said that an effort was made to get "typical" or"representa-
tive areas. Another practice is.to make surveys where they are asked
for or needed. '

Qisadvantages of survey method: Those mentioned were:
1. Not enough detail for some types of problems. v
2. Not accurate enough for close analysis.

3. Not large enough sample for some purposes.

Accuracy of data:

Mention was made of checking of survey records against route
records, creamery and elevator records and store records and
finding them reasonably accurate; but no actual evidence was
furnished although it was asked for.

As to bias in answers to survey questions,' mention was made
of the danger of leading questions, and of the bias of some field
men. One person who had worked on cost—of-production supveys
reported that certain of the field men nearly always brought in
the low-cost farms and certain others th e high—cost farms. Where
considerable experience has been had in survey work. the feeling
seems to be that field men can be sufficiently well drilled to
obtain detailed information accurate enough for most practical
purposes. Whether this feeling is justified of course depends
upon the sorts of purposes that are deemed practical. For ex-
ample. will surveys provide data accurate enough to show the
most profitable rate of feeding?

‘ .
' 5

Farm Management Surveys. (1)

3:1.

One of the purposes for which surveys are frequently made in
farm management research is to obtain a picture of the farming of a
region, - the size of farms, the income and expenditures of the farmers
the different types of farming represented in the region and so forth
to show the general agricultural economic situation in the region select-
ed. Another use to which farm management surveys may be-put is to
show features of farm organization for different systems of farming in
the same area. When such is the purpose, the make-up of the schedule
and the selection of the cooperators will be different than when the
purpose first mentioned is intended. , ' - *

Egg ' ' Efficiency in the survey method is gained by 0
Blank the use of well prepared printed field sheets.
Such field sheets save time and result in q;
- greater-accuraCy; Mimeogrsphed field sheets
' ' »‘ are less costly when small numbers of records
are to be_taken, but the bulk-of the'mimeographed sheets and extra
space required for filing are sufficient disadvantages to Justify the
expense of_prihting the field sheets-whenever a'few hundred records are
to be taken." Occasionally the same.form is used for both Office record
and field sheet.‘ This however has the disedvhntage of not furnishing
enough free Space fbr making'notes or chlculations when the record 137
being taken. ‘ -‘ 4

Ordinarily records taken in the field should be copied at
once\onfonns that_are used in the office. It is very important that
these forms be printed on a heavy linen ledger paper to stand the wear
and avoid wrinkling that may come from use in the office; ' ‘

Office forms should be in much detail, and provide for
plenty of information. In preparing the blank, sometimes forms may
‘ be cut and pasted together,and several trials and adjustments shouldv
be made until the best form and arrangement is reached. Some time
spent inzfirranging the blank, deciding on the width of the columns to
be allowed, and the arrangement of the informstion will facilitate
transcribing,completing and tabulating the record. Copy sent to the
printer should be of exact size required, should be written with indie
ink on one side only of heavy manilla paper, and spaces indicated ex-
actly as desired on the printed'form. Copy'for each page of the
form should be prepared separately; Attention to some of these de—
tails will result in a good blank.

 

(x) Abstracted from paper by Professor E. G. Misner, Cornell university,
avoiding duplication with report upon the inquiry.

H

' 64..-,

sMaps should be provided for all areas surveyed. Soil maps

'showing the contour lines are most useful in farm management surveys.

On such a map can be left a permanent record of the location of each
schedule included in the study.

Organization ' The number of enumerators desirable in
for field ‘ the field party or crew.depends upon the
work type of schedule, the total number of

cooperators to be visited, the'distance
between cooperators, or the proximity
of the farms to some convenient community center. Ordinarily three or
four men to a car prove an efficient field crew for most types of farm
management survey work. When studies are made of businesses that are
Widely separated and require more travel between, one man with a car

‘53 more efficient.

Whenever a large field party is to be used in a survey, it
is very important that the enumerators be well drilled and trained in
the taking of records before starting work. This means as a very
minimum that they should have taken at least one record in drill
work, made the trasnfers to the office sheet, and completely worked
up the office form so that they will be familiar with what is de-
sired. Meny enumerators have done survey work in the past with even
less preparation than this, with such results as one would expect.
The one who is to write up the results or who has had experience
in survey work or preferably both should be in charge of the party
and check the work. The drill work should take up such points as
the way to approach the cooperators in an easy solicitous style,
the influence of the method of asking questions on the bias of the
answer, and ways of detecting errors in making enumerations.

Records should be checked at once for omissions and mistakes.
When more than four men are in the field, one man should usually stay
in the office, plan the work, Check the records and keep things moving.
If the information needed is not obtained” the enumerator should be‘
send back to the cooperator the next day,provided the omission or
error is of sufficient importance to justify the expense and trouble.
If it is only a minor point, it may be obtained by correspondence.
Some system of checking records in the office, such as attaching a
sheet of paper with a notice indicating the omission or errors in the
field sheet, and checking these off one by one with the enumerator as
the corrections are made, should be followed.

Suggestions The cooperator should be approached in
about asking an easy style. Every effort should be
Questions made to get him in a state of mind and

so situated that good work may be done.

If others, su.ch as neighbors, are present,
it may be unwise to proceed with the record. Generally it is desirable
to avoid the family circle, altho often other members of the family may
be able to assist in giving more accurate information. Usually, how-
ever, they interfere with the facility of taking the record and its

-_- ‘_ .
,I

65

accuracy:“" Enumerators should take plenty of notes. A definite system
of signs and abbreviations should be adopted to help the enumerators
proceed more rapidly with the questioning. In all survey work, the.
rules of good bookkeeping should be observed. Decimals and figures
should be aligned. This is mostly a matter of habit and can and will
be followed on the field form as well as in books or on office forms
by those so inclined. Persons careless with figures had better choose
other fields of work. '

x

The order of questioning need not necessarily be the ordor‘on
the blank. A good enumerator will soon develop the knack of taking
down the information from much of the conversation that is carried on.
In fact, the more nearly one can get away from the cold formal method
of blunt questions, the more successful one will be in field work of
this kind. The best time to approach a farmer is early enough in the
morning so that he has not statted the morning field work, or Just
after dinner. '

Questions should be stated directly and specifically to speed
up the work. The enumerator should assure himself as much as possible

that correct answers are being given, that the farmer is neither trying -

to make too good an impression. nor to understate his earnings. Tables
of supplementary information should be carried in the field to assist
enumerators - silage tables with the capacity computed for various
silos. and so forth. Livestock records should all be made to check as
to numbers. However, one of the most important things influencing the
accuracy of a survey record is the judgment which the enumerator uses
as to the amount of time to be devoted to different parts of.the record.
It is unwise to spend too much time on inventories. One had better
spend that time in obtaining expenses accurately from books or in get—
ting the receipts more accurately - items which will have a decided
influence on the net returns of the business.

When a study is being repeated in the region, ending in-
ventories for the previous year will serve as the beginning inventory
of the succeeding year‘s record. This makes it necessary to obtain
only the and inventory in a continued study. The depreciation on
horses or machinery as well as the changes in the_inventories of
cattle may be obtained in this manner. Market veins changes should g
be obtained, and indicated separately on the record, but not included
in the computation of the net returns for the business as a whole or
for the enterprise. '

Prices and values should be obtained for each farm.~ The
supervisor of the party should watch out for the tendency on the part
of some of the enumerators to get prices and values for one farm and
insert this on all succeeding records. This destroys one of the main
purposes of the study, namely, to get a mean value for various items
on the schedule. To do this accurately, replies from all cosperators
should be obtained. The enumerator should guard against the thwarting
by the cooperator of his efforts to obtain such figures. He should
stick until the last goose is hung. and avoid making any leads or
suggestions that will be likely to result.in an answer which was;

‘1

66'

really suggested.by the enumeretor.(x)

Organization When the records are finished in the
for Office ' field, the next step is to check in _M
Work the office the parts of the record whtCh ‘\

are necessarily compensating. If the

cost of a new barn is included under
farm expense, part of it if not all should appear as an increase in
capital. If, on the other hand, a barn was;destroyed by fire during
the year, the insurance received may appearfiunder miscellaneous re-
ceipts. but it should also appear as a decrhase in the farm capital.
.This means that usually the person in charge of the survey must check
‘the records in the office for offset items of this nature. If any of
the office force are sufficiently familiar with the work and the nature
0: the items to be checked, it may be desirable to trust it to them.

After the work is thoroughly and completely checked in this
manner, the extensions should be made. It is desirable to use a diff-
erent colored ink for such work. This means that less of the check-
ing will be overlooked. The work of one person in the office should
be checked by another. It is always a good plan to keep the less ex;
perienccd clerks at calculating and trust the checking only to those
who are more experienced. Some have found it best in farm business
analysis work not to complete the financial summary of the record
until some sections of th record have been tabulated, especially
those sections where a slight mistake or an error would directly '
affect'the returns. This may save some tine in enusing the figures
and making corrected summaries. Two kinds of checking are always
desirable in work of this kind - inspection"hecking and machine
checking. Most persons are fairly good atEachinc checking. The
ability to detect errors by the inspection ' the data is less well
developed in all of us, but an ability that hould be encouraged
because it is one of the best means of expe ting statistical work.
Just running over the data in a thoughtful _' will often locate the
mistake in a tabulation. Misplaced figureagand misplaced decimals
are the most frequent of all errors. Many bf these can be very
easily detected by inspection checking. .

aniliarity with short cute is desirable. Many persons
are still calculating interests at 5 per cent by multiplying by .05,
instead of the easier and more rapid way of dividing by two and
pointing off one place. ‘

Making ' After the records are completed and
Tabulations ready for tabulation, the first thing

of importance is to decide on the re-
ocrds to be included. Any incomplete
or freak inaccurate records should be

 

(x) The question has been properly asked by members of the committee
as to whether figures obtained in this way have enough validity
to warrant getting them.

67

omitted. Considerable care should be taken in deciding on the schedules
to include in the tabulation or final summary. Once an undesirable
record is included, the averages of various groups in the subsort prob~
ably will be greatly disturbed by this one record.

The next step in making tabulations is to decide on the. sorts
and subsorts to be made if the cross-tabulation method isto be used,
or on the items to be punched on cards, if this method is used. Ordin- ~
erily most farm management surveys should be tabulated completely in
one or possibly two sorts. A trial tabulation should be run bifore do.
ciding postively on the main sort to be made. , ~ '

After the sorts and subsorts are decided upon, the tabulsp
tion sheets should be prepared or the cards laid out for punching;
Prior to the preparation of the tabulation sheets, a list of items
to be tabulated should be prepared as a guide. Proper headings for
each occurrence should be left on the tabulation sheet. The headings
should be in the same order as they appear on the blank. They should
be properly ruled and underscored to indicate the cblnmhsto which each
heading applies. Doublecolumnsfbr loss or gain. or for-plus or minus
. quantities, should be carried to faciliate the summarising of the date.
A good margin of at least two inches should be left at the left of each
tabulation for binding. Tabulation sheets should ordinarily not be
spliced, but the tabulation carried over to-a new sheet on which the"
heading should be repeated. In. drawing off the figures, the trans-
fers should be aligned in nest, clear figures. Cerelessness in tab-
ulating is costly and is to be condemned from every standpoint. When
the value is zero, a cipher should be put in the column; When there
was no report, it should be so indicated by a dash.

'Report of Tests of Survenyaterial.

nggling ' Table 1 following shows what is gener-
ally recognized, that the farms included
in surveys tend to be larger than the
average.‘ In the South Central group,
this is accounted-for partly by the in-
cluding of plantations as farms in many cases; in the Mountain group,,
partly by the fact that ranches have been seleCted for-survey study
more often than farms. (Sugar beet farms have been omitted from ‘
the table). The differences in the other’sections are about the
”usual ones. The New York state surveys average a fourth larger
' than the census figures. ~ ’

- \\B-' 0,. r-H'.

Tablevl.» Comparison of Averages of Survey Farms with Those /;’
of the Census for the Same County at Nearest
‘A-N Censue'Date.

I

 

 

 

: Number : Number : Average Size_of Farms

: of : of : .-

: Farms 3 Surveys(t): Surveys (2) : Census (3)

Total ; 36879 i 379 i 223 ; 148‘

>39w England ; 3317 ; 49 g 151 g 105
Middle Atlantic ; 8595 t 49 t 120 E 94
East North Central 2 5252 E 78 ': 141 : 110
west North Central ; 6504 i 76 K: 347 ; 265
South Atlantic ; 5474 g 50 i 149 i 86
East South Central ;’ 1215 ; 11 l 267 g 97
west South Central ; 1173 i 11 E 316 h 117
Mountain ; 2736 g 27 ; 315 ; 269
Pacific 2 1713 i 28 ; 201 .;. 194

 

(1) Surveys of highly specialized farming, as of sugar beet growing
in an irrigated section, were omitted.

(2) U. 5. Dept. of Agriculture Yearbook 1925, Table 652.

(3) 1910 Census compared with surveys in 1910-15.
~ 1920 Census compared with surveys in years 1918-20.
1925 Agricultural Census compared with surveys in years 1921-24.
Census figure used was the average size of farm of the county
in which the survey was taken.

69

Tame LL. Comparison of Values of Farm.Propert;r of Survey
Farms with @1080 of the Census of the Same

Cmczfly-‘M the Nearest Census Date.

Averégc Value of All Farm

 

 

“Kare. ' : 130. of Pro-par by - Pe'r Farm
: Inclusive ': Sufvcys 2.”. Survey : Census (X7
New England § 1997-1915 § 35 : $ 7,986 § $ 4,742
Q 1918-1920 Q 2 '2. 14,517 2 9,332
: 1920-1924 : 12 f 11,908 i 8.620
Middle/Atlantic 1 1907-1915 : 37 : 9,584 : 5,910
‘ : 1018-1920 : j2 : '18,188 Q 12,918
E' 1,21-1924 9 .5 § 14,044 _; 8,679
East North céntral I 1907-1915 g., 48 ; f21,877 ; 11,814
} . ; 1919-1920 § 15 1 26,757 Q 22,284
, : 1921-1924 2 20 :7 27,954 Q 13,424
What North 0e8tra1 § 1907-1915 : 54 :' 23,531 1 14,937
’ ‘ Q 1918-1920 Q :6 § 52.022 Q 35,452
Q 1921-1924 § 16 § 41,841 2 27,408
. ‘? f 3 3
South Amlant1¢ Q 1907-1915 § 13 § 12,925 Q 3,877
‘ ; 1918-1920 ; 19 : 15,179 g 4,809
§ 1921-1924 § 9 § 9.159 2 5,408
Best South Caitral 2 1907-1915 Q 3 E 33:655 : 4,388
E 1918-1920 ; 6 § 28,163 ; 3,758
Q 1921-1924 ; 1’3 § 16.469 Q o5,994

r...-

70

Table II. - Continued.

 

Average Value of A11 Farm

 

 

 

gYears ; No. of ; Property — Per Farm
: Incl...ive : Surveys : Survey : Cenwsfl)
: : : z ._

West South Central 3 1.9ow-1915§ '7 Q $ 18,811 Q 3; 5,751
Q 1918-1920; 2 - E -

Q 1921-1924: 2 27 .793 Q 6, 521

~ : : 1- :

Mountain : 1907—1915: 19 : 13,793 ; 8,807
2 1918—1920; 5 Q 38,008 1 23,600
; 1921—1924; 4 ; 26,792 ; 10,921

ggcific 1907—1915: 10 16,928 11,057
; 1918—1920; 1 : 51,200 2 51,810
; 1921~1924; 5 ; 22‘116 : ‘ 12.254

(x) Same comparisons as in Table I.

\
o\

71

In Table II, the same comparison is made in terms of values
of farm property, with sub-groupings according to the nearest census
date so es to make allowance for changes in the price level. The
discrepancy is much more pronounced. (That for the 1907—15 surveys
is partly explained by the fact that most of the surveys were made
in 1914 and 1915, and land values had risen a. good deal since 1910).
This suggests that the fame covered in the surveys in many areas
are better farms with probably 9. larger percentage of improved land.
Table III gives the same date. in more detail for three such surveys.
In Clinton County, Indiana, the survey and census figures were
closely alike in most items. In the Virginia county, the differ-
ences are very large for most items. ‘ Difference in system of
farming seems to ~thigure in the Delaware comparison; although the
survey farms are practising the more extensive farming, their land
is the more valuable. . '

I

Table III. -

Comparison of_Survey Data with Census Data
for the Same County in the Same YBar.

 

 

 

50793'“"

1- : : :Newdastle,fiel.
1 :Clinton 00. 1nd,:Frederick Co.Va.:Middletown Area
- : 1919 : 1919 1 1924.
_’; :Survey : Census: Survey : Census: Survey J:Census
: : i 3 : 1
1-1.. , i x : : : ‘ : _. .
N0..of farms 1 100 : 2399 : 125 : 1725 : 55 :;,1957
. Acres per farm : 130 z 104 : 159 : 133 : 215 : T107.
.¥ 1 : : : 2 2
Céapital per farm - Total :332318 :$27037 :$33676 :911151 :$22832 :912785
Land '2 25573 : 21074 : 25301 : 7433 : 13360 : 5715
guildings : 3129 ; 3054 : 4010 : 2224 : 4944 :
Maphinery : 484 : 976 : 588 : 545 : 1105 : 94?
Livestock : 2575 : 1933 : 1572 .: 959 : 2432 :{,1044
Iarm supplies : 750 : - : 334 : — : 649 2" J
gash to run farm : 106 z — : 821 : - : 342‘ 1 ~
Yieid per acre : : : : : . :
Corn bu. : 51 : 47 : 34 : 32.3 32 : 30
Wheat " : 42 : 42 : 14 : 12 1> 23 : 21
eats " : 20 : 20 : 22 : 17.: ‘ r :
Rye " : 15 : 15 : ll : 8 : _ E :
Hay tons : 1.0 : 1.1: .9 : 1.0: ‘1.1 : 1.2
Acres in - : : : : : ,:
Oorh' : 45 : 31 : 16 : 12 : 38 '1 15
Wheat : 14 : 15 : 23 : 15 z 67- : 19
Oats : 22 : 15 : 1 : 2 : - ' :
Rye : 2 : 1 : 1 : 3 : - :
gay : 29 : 14 : 15 : 9 : 25 : 15
, : z : : z :
3 : : : : : t
.' 2 2 t i 2 3
’ : : t : 3 1

 

73

The most interesting comparison is that between the 1115 ;
farms recently surveyed in North Carolina and the census average ’
for the state in 1925 sflbwn in Table Iv. This survey was planned
to sample the whole state. First the state was laid off in four
districts, the Tidewater, the Coastal Plain, the Piedmont and the
Mountain regions. Then either three or four counties were marked
off in each region as presenting the best examples of certain sets
of farming conditions. Within each county, the sampling was done
by sectioning. The survey farms are almost twice as lemge in
total acres, although only a half larger in crop acres. In value
of real estate per acre, the two are nearly identical, but the
survey valuations of buildings were considerably higher per acre.
This is probably due more to the differences in the two methods
of getting the valuations than to anything else.

74

Table IV.~ Comparison of 1928 North Caroline Survey Data
and Questionnaire Data with State Census

 

 

Averages.

: : : Federal

:1 Census : Survey :

Q : : Questionnaire
number of farms — ::QBQy884 : 1115 : 380,

81 e : :
Crops :t 24 : 37 z -
‘ '5 : :
Pasture - cleared :1 5 : 9 : -
“ 2, : :
Pasture — wooded :, 4.8 : 10 : -
Woods : 30 z 54 : .
Other Land : '6.5 : 18 : a.
All land : 65.6 : 128 : 157
Land : $ 2.421 : $ 4,600 : (

: : : ( $ 6,931
Buildings : 845 : 2,059 : (
Livestock : 273 : 404 : 514
Machinery :. 163 : 190 : 348
Feeds ‘A - : 167 : 344

:r : :
Total x: 3,704 : 7,430 : 8,137

:i : :
Farm Ezgenses: :3 : :

: (1) : :
Feeds : 80 : 71 : 82
Fertilizer : 134(1) : 203 : 188
Labor (Wages) : 116(1) : 124 : 296

 

(1) Per farm for the farms reporting such expense

75

The comparison in Table V is between the survey averages and
the county averages for each of the four regions. (The questionnaire
data will be explained later). The counties chosen average 65.9
acres per farm as compared with 65.6 for the state as a whole“ Their
land averaged $51.53 per acre in value as Compared with $49.80 for
the state. In the Mountain region, the survey farms were the more
valuable per acre, and in the Tidewater region the "everse was true.
The difference in size of farms was greatest in the Caestal Plain
Region. ‘ ‘ ~-

Table V.— Comparison of Survey and 1925 Census Real

Estate Values for the Same Counties

by Districts of North carolina.

w

 

 

 

o
v

: Number of : Acreage : Value of : Value of
: Farms : per Farm :Real Estate : Land
Districts : : : per acre : per acre
: : 1925 : : 1925 : : 1925 : : 1925
z:Survey:Census:Survey:Census:Survey:Census:Survey:Census
: 1115 : 42541: 128 : 65.9 :$52.10:$51.53:$35.94:$38.88
~'L I t v I 0' 0 t 0
'Coastal Plain Region: 294 : 15382: 136 : 56.6 : 65.70: 65.95: 45.06: 45.94
Mountain Region : 281 : 9129: 12:5 : 76.4 : 40.80: 54.45: 30.02: 26.69
Piedmont Region : 311 : 9912: 109 : 74.2 : 55.85: 50.71: 36.34: 37.40
Qidewater Region : 229 : 8118: 150 61.5 : 43.82: 51.45 30.81: 38.78

 

(x) This survey was conducted jointly by the North Carolina

Tax Commission, the North Carolina Experiment Station
and the U. S. Bureau of Agricultural Economics.

'7'?"

In Table V1.13 presented a. frequency distribution of farm
acreages according to the three systems of collecting data. It is
the small farms that are omitted, especially the very small farma,

in the surveys. ‘

Table VI.— Frequency Distribution of Size of
Fame in North Cerblina Accord-
to Three Medthot! cf Collect—

 

 

 

 

ing Data.
.1-
Manner ’of Farms 2 Percentage
: > ‘. :Question- :. : ' . :Question—
: Census : Survey : naire : Census : Survey :. naire
Total £‘283.284; 1115 Q ‘ 380 Q ”100 Q‘ 100.. 1-4100,
Under 20 acres ; 55.047; 41 Q' 10 E 23.2 § 3.7 Q 2.6
20 - 49 ; 93,817; 214 Q 43 § 33.0 § 19.2-: 11;3 1
50 - 99 ; 58,1291, 333 i 91 : 24.0 2 29.5 ; 23.9
10029174 2 ’37;611§' 313 Q 125 ; 13.2 ; 28.1 E 33.1
175 -259 ; 10.543: 115 Q 54 i 3.7 Q 10.3 § 14.2
250 -499 i 5.525: 59 ‘i 45 i 1.9 i 5.2 i 12.1
500 and over. ; 1.710 ,.30 ,; 10 Q .5 E *2.8 E 2.5

o. .- on go no

 

- ”a... .. - -. . - _ — ._ ._,._..__--.. .

78 "_ ‘r\_
a \; \\‘
This raises the significant question as to shnther the
census definition of a farm is suited to the purposes of'guch of
the analysis of agricultural problems. ‘It is aimed largely at
securing a full count of the crop acreages and production, as no
doubt it should be, rather than at describing farming itself. In
most bf these surveys, no conscious effort was made to get farms of
larger than average size. Instead, everything that looked like a
real farm was taken as it was reached. The farms omitted were
mostly those of persons who earn much of their living at something
6180. In the mountain regions, many of such flamilies were living
on a largely self—sufficing basis; in the Southern states, largely
in the same way, or by occasionally working out. ‘The census fre—
quency distribution therefore includes a large group of people who
do not stand as farmers in the eyes of those familiar with the prob-
.mlems of farming. The survey distributions may much more nearly
‘approximate true agriculture.

The lesson to be drawn from the foregoing is not that the
survey distributions are not fair samples, but that we do not know
in most cases whether they are or not. A more conscious effort
must be made in advance of the Survey to define the universe exactly,
to 33y f0? each31rvey exactly what shall be included as farms and
what not, and to publish these specifications in the reports. of the
surveys along with information to show something about the farms
that have been omitted. It would be well to collect a little in-
formation about the farms that are omitted - acreage, principal
sources of income. both from the farm no outside, etc.

The conclusion is obvious from the foregoing discussion
that present Sampling procedure in survey-taking is not altogether
satisfactory. The method of working out from some central point
until-a certain number of records has been taken, results in getting
too few farmers living remote from town. It also raises uncomfort-
able questions as to whether to work another day or two and reach
out into conditions apparently different from thOSG encountered thus
far in the survey. Likewise the method of following each of the

V main roads out from town fails to get enough of the farmers living
on cross roads. There is alwirs danger in omitting farms which
do not seem to promise a good record or which are abnormal in any
particular.

The proper sampling procedure depends upon the objective
of the survey, principally upon whether its purpose is to obtain
data describing conditions in a large area, or instead, to study
the relationships existing in certain type of farming. If the
objective is the former, it will be well to take definite geograph—
ical units such as townships. If a larger area than an single
township is wanted, additional townships can be taken according to
some checkerboard arrangement. Another'possibility is to-aelect
sections (640 acres) checkerboard fashion. A still more satis—
factory procedure, except that it involves more trouble than the
others, is to check off every 5th, 10th, etc., farm from a complete
list of farms in the region to be studied.

It is not necessary for such an objective that the study _~
~Ebouldeinclude every type of farm in the region, It can be defin-
italy agreed in advance that rural mail carriers and others earning
most of their living from other than farming operations shall not be
included. In studying farm families it can be agreed that bachelors
will not be included. ‘

If the objective is of the second sort, namely, to study the
relationships in a given system of farming or set of conditions, the
important consideration is to locate definitely in geographic terms
the area in which this system of farming occurs in the form'desired,
and then Sample this geographic unit; It is entirely proper_to mark
off a small area on the map_and to include every farm in it except
those which are discovered in the course of the survey not to be prac—
tising this type of farming. . When the results of the survey are
published, it should be made clear that they relate to the situation
as found in this limited area. There should, however, be enough
study of the larger area from which it has been selected so that safe
statements can be made as to the more general application of its con-
clusions. V . :

The method of selecting areas on the basis of some plan of
stratification according to conditions, is almost necessary in certain
types of study; for example, a study of land settlement conditions in
a region so large as the Great Lakes states. It calls for a very
careful preliminary study of existing data with respect to the region,
and probably, in additionga reconnaisance survey; and-all the differ-
ent conditions and combinations of conditions that occur in the whole
area should be listed and placed on the map. Proceeding along these
careful lines, it is possible in most cases to sample a large area
without any bias, and to include all the different conditions in some-
where nearly the proportions in which they occur in the region. It
is not altogether necessary that the proportions shall be the same,
but if they are not; an average or summary of the figures must be
properly weighted. ' '

Samples need to be large in area studies in order to give mi;
all the various conditions existing in them their proper chance of
being included. The various tests of sampling, such as probable
error of the mean, are not likely to be a very safe guide when ap-
plied to most of the data in an area study, because the taking of
the cases does not comply with the first condition of simple sampling,
namely, that all parts of the record shall be included in each sample.
What happens instead is that as the sample is enlarged, the field man
reaches out into new sets of conditions. ‘

The practice should be followed more generally than at
present of having the field leader test out the sample as to size
as the survey proceeds.

so .1

/

“NM , When the records for the 800 iwflwfi;
§\ ‘ in Vinton‘\Jackeon and Meigs counties in Ohio ,
\<; in 1926 hadfbcon worked up in the office, it
\ seemed that too large a. proportion of them
'"77 " " "N'. showed consiaerggi: larger cash outgo than
°35h~1n°°me. Accordingly 73 of the fami s were ro—surveyed; that

is, the same field record was taken back to fire farm and checked over
item by item. Table VII shows the results of thim- ~survoy. Of
the '73 farms, 40 had their balance of income over outgomeised an
average 'of $375. Some of the miscellaneous items were board and.
lodging, interest on government bonds, dividends from stocks, receipts
from operating a bus, notary‘s foes.- The extra poultry receipts were
largely from turkeys, most omitted in the first survey.

Table VII. Results of Ila-survey of 73 Ohio Farms.

1"

 

 

 

: Decrease ” InoreeSe3

:No. : Total : :No. :Total : .

:of : :Avera’gpxof : :Averagc .

:Farms:Amount : :Farms :Amo'unt : ' "

. 3 : v t ’
Cash income : 1 : $120 $120 : 38 :$15, 059 : $396.20
Cash outgo : V, : 135 : 33;?5} 2 : 285: 147.50
Cash income less 011th : 1 : 120 : 120 : 4O : 15.008: 3715,20

 

Sources of additional income.

No. Amount

Work off' farm 14 $ 5.960
Insurance 3 _ 8,230
Pension 3 1,440
Cattle 5 1,127
Coal mined on farm 2 1,000
Poultry '7 759

Misc. 5 1,068

81

How a re —survey bf the rest of the zoo farms would have
changed the totals can only be conjectured. It would appear from
the nature of the changes in the 73 that it would have raised the

cash receipts of many of the remainder, though probably nowhere
nearly so large a proportion of them. No doubt also some of the
larger surpluses of receipts over expenditures are due in some
measure to omissions on the expenditure side.

The omissions found in the Ohio ro—survey were mostly in
the group not classified as_farm business income, with which most
field men have had little experience. No doubt Mr. Hefithorne's-
long experience taking strictly farm'business surveys may have in- /
clined him to overlook this other group. It must also be added
that Mr. Hawthorne is one of the most experienced and most careful '
field men in the United States. ‘He has worked on many thousands
of survey records in the office and knew in advance what errors,
field men make on the regular farm business schedule. The majority
of field men are far less prepared for the regular farm business
survey taking than Mr. Hawthorne was for the work on other sources
of income. v

In surveys of several farm families made by Dr._C. C.
Zimmerman in Minnesota in 1925 and 1926, in which tote; receipts
of the farm family from all sources were bilanaeed against ex-
penditures of all kinds,=rven though for single years, only a.small.
percent showed expenditures appreciably in excess of receipts, al—
though several of the areas were depending upon cash crops to a con-
siderable extent. The balancing was roughly done in the field when
the schedule was being taken, and checked over carefully later. Follow—
ing is a clussification of the reasons for the excess of expenditure
over recei t in the 56 cases. The excess in mest of. them was rather
small. TEe agricultural depression figured in many of them:

Farms still in process of development......................;.........10

Interest on heavy mortgages and poor or only fair incomes...... ..... .10
Crop failure or low prices or both, and inability to reduce

expenditures to low enough level.................. ....... .........10
Purchase of automobiles, trucks, lighting plants, or house

improvements, out of One year's income..... ......... ..... ..;.....14
Wheat crop held unwisely............................................. 3
Slump in prices of purebred cattle............L..................... 3
Extravagant living, not yet adjusted to lower income....... ...... .... 3'
Miscellaneous.................................................;...... 3

It seems apparent that the checking of complete farm family
expenditures against complete receipts would aid to the accuracy of
the farm business data and would be highly desirable if it did not
make the schedule so long. .

No doubt the item which one would expect to get most accurm
ately in a survey is taxes. Table VIII summarizes a comparison of the
survey returns with the reports in the auditor's books for the 1115
North Carolina farms. The survey reports run about 5 per cent higher.

N . ,
82 " ___,\ 1

\

_‘ A total of 405 farmers reported taxes averaging $ 25.58-mcrev and a

total of 224 reported taxes averaging $19. 90 lower. It is signifi-
cant that 233 were too high or too low by 20 percent or more, and
157 by between 10 and 20 percent.

Professor G. W. Forster makes the following report upon '-
these discrepancies: -

1. In most instances the farmer reported with his tax the poll
tax and dog tax while the auditor gave us only the tax on prop—
erty. In checking over the results in.tuh ecunties, we found
that by elimating the poll tax and dog tax that the differences

\ were reduced from 5 per cent to 2 percent.

2. The land upon which the farmer reported tax was often different
from that reported by the euditor.-e

3. occasionally farmers pay two years' taxes at the same time;

4. In seme cases the farmers reported the taxes incorrectly. For
example, in one inStance which I have before me. the farmer re- '
ported $133 in taxes when the auditor reported $99. In explain-
ing this. the farmer stated that the $133 was. for 1926 and the
$99 for 1927.. .

5. A drainage tax was often reported by the formerr

6. 'In a few cases the records were taken from individuals who
knew the farm business but were not informed in regard to the
tax.

The accuracy of the survey results varies according to the
education of the darmers and their practice in the matter of record ,
keeping. In_many areas, relatively i‘ew keep records of any sort.
In Hillboro Cdunty, Florida, 44 out of a hundred,_representing 54?
percent of the farm recei.pts. kept records of receipts and expendi-
tures; in Polk County, Florida,. 53 out of a hundred, representing
64 percent of the crop receipts... These farmers were fruit growers
in Frederick County, Va., 28 kept records.' These are recognized as
exceptionally high percentages. ‘ I ,

There is need fdr more concern over the accuracy of survey
results than is géherally‘felt.. .The tendency is to assume that since
the method has been widely used it is dependable. There is abundant
most favorable conditions. Granted that the method is dependable used
by institutions with long experience with it on standardized schedules,
this does not mean that it is when used in other institutions or in
new fields of research. Inexperienced field men always let many
mistakes slip into their records. More attention must be given to
drilling field men in advance. They must be watched closely to see
that they do not ask leading questions,'end that they do not handle
the questioning in such a way as to let personal bias influence the
answers. The leaders of field crews must cheek the schedules more

86

Table 1x.- The Deviation of Farmers' Tax Reports from County Auditors'

Books -— North Carolina

 

: : Dev1atron of”Fhrmars' Reports from Counfiy Audltor's Books
:Average: Farfiur storfs I‘Eeiow :: _—-.—Ffi?mer FEfifiFFE — A ove
: Taxes :NBTEf: T5$EI"‘TKVErag8:" ::N0.of: TFfEI‘Dé~: IVEFEEe : Ear-
County ': all :Barms:Deviatn :Deviat—zLar-::Farm-:viati0n :Deviation : gest
;Farm8rs:Rep0-:jon for :ion for:g8st:: 8rs :for Fann- :for Farm- :Deviat-

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. :rting:Farmers :Farmers:p8v—::Repou:er5 Repor-zers Repor-: ion
' : ':R8port- :Report~:1atn::rt1ng: t1ng : ting :
f f 5 15g .5 lng .3 53 3 5 E
Total 23101.51; 224 i_§4,459 3 $19.90; 23 405 E 810.361 3 ‘$25.58' E
Ashe § 110.189 .22 i 187 § 8.505 838:; 40 i 726 in. 18.10 §_8118
ngden g 128.11§‘ 10': 534 g 53.40: 203;; 10 i .268 E 26.80 E' 10
Qgfigwba g 75.19; 23‘: 267 § 11.60? 64;; '25 3 204 i‘ I 8.15 E 25
gggberland § 117.19; 24 § 555 :f_§§.12§ 243?; 24 i 1.569 E 65.37 E :24
Davidson g 56.42§”'35'§ 434 i;;12,40§v 62:; 31;? 715 i 28.06"§ '81
Hoke § 114.605 5 g 119 § '23.90? '65:: ' 9': 79 g 8.77 g 9
Jackson g 68.91; 10 E- 108': 10.80?‘ zoiél 27 § 232 Z . 8.59 E 27
ngoir g 185.63;__10;§ 1.069 § 106.905 573;: ‘48 i 1.554 g 32.37 i 48
Méggn g 62.10% ~ 0'; _ o i '- 0?" 0:5 0 j. 0 i‘ 7 0 g 0
McDowell g 65.01? I 6,: 44 i 7.33: 203? gng' 645 g 20.17. § ' 82
MQQre g 88.30%_ 11'; '171 §“'15.50§ 915; .13 i 110 § 8.46: g 13
EfiggflL § 277.305' 12 E 150 :fi_12.50§ 65:: 6‘: 218 § 36.30 g '5 6
Eergumars 1;;1oe.62§ "7 g 83 §' 12.42: '31:? 9 I __187 ifiigxgeo g' 9
Egrsons g 82.14; 9 E. 90 g 10.00: 26:: 47 E 925 g '19.68 g 47
Bgnder g 85.87; 23 i 895 § 17.07? 176:: 50 E 958 i 19;06 E 50
HEEflEL Efilgggeog 9 § ‘ 100 § 11.11: 40:: 14 E 213 § 15.21 i 14
thwan 2 166.80; 20 § 149 § 88.62: 51;; 20'; 1.768 g 88.15 i .J20

 

 

%

 

H

8’4

»‘rigorously, and compare the returns of the different field men. Doubt-

‘ful interpretations or a_v?lications of questions- -should be settled once
for all as they arise. ind adhered to thereafter. Less stress must

be placed on number of schedules aer day and more on accuracy of re-
sults. Some field men become "record hounds".A11 these precautions

are hivhly irportant with new types of surveys, espe cia1.ly these re-
lating to interpretation of questions and instructions.

The accuracy of survey data, however, is largely a matter of
the nature of the data. The accuracy is relatiVely high on such items
as taxes and major cron receipts and major expenditures, and relatively
low on a large list of minor items. No doubt much individual bias
enters into questions involving judgment. Consequeitly, the further
d.iscussion of accuracy is best taken up in the next section connectiOn
with the different types of questions.

Analysis There is no doubt that questions can be asked
of ass of which call for a definite answer that will be
Questions: very nearly correct. The following questions

proved, upon use, to work out this way: The

first-payment on his farm by the settler in
the cut-over region. The amdunt of his cash on hand at time of settle—
ment after making the first payment on the farm. The amount of money
contributed to church during the year. The amount of the average weekly
meat bills. The first two of these represent events standing out very
clearly in the memories of the persons involved. Church_contributions
also seem to make a definite impression on the brain substance. -No
doubt in some cases the amounts were calculated on the basis of weekly
contributions of small amounts. 3 table-of weekly meat bills shows
that these were frequently estimated rather closely. Most housewives
will be able to make a very good estimate on such an item.

 

 

Frequency Distribution of Frequency Distribution of
Church Contributions of Average Weekly Meat Bills
125 Families: of 190 Families:
Nothing 42 families $ .75 4 families,
$ 1.00 8 410— 1.00 14 ~do—
2.00 20 -do- 1.50 19 -do—
3.00 11 —do— 1.75 8 ~do—
4.00 11 ~do— 2.00 21 —do-
5.00 21 ~do- 2.50 20 ~do-
6.00-10.00 6 -do- 3.00 24 —do—
11.00-15.00 2 —do— 5.50 17 ~do—
16.00—20.00 2 —do— 4.00 19 ~do-
Total 123 families 4.50 9 —do-
5.00 12 —do—
5.50 3 —do-
6.00 8 -do—
6.50 3 —d0—
7.00 4 -do— _
8.00 2 —do—
9.00 2 -do—
9.50 2 —do—
12.65 1 —do-
17.43 1 -do—
11.00 1 -do-

"" fit? 190 F'smi 1": es

"’85

, A question of somewhat similar nature that obtained useful
and significant results was the following: "How much has the sale
price of wild land risen immediately around you since you settled
here?" There were 48 settlers who specified definite amounts ranging
from $2.00 to $25.00 per acre, about an equal number who specified no
increase at all. 10 who said that it had doubled. and 10 who said that
they did not know. Upon correlating these answers with the number of
years that the settler had been in the new country, the-curve of in-
crease in wild land checked very closely With the price actually paid
by the settlers locating in the same areas. The related question:
"If pieces of wild land are now being held for sale in your neighbor-
hood, what is the price?". gave equally satisfactory returns, except
that there was more of attendancy to state a range rather than a
definite price. It was found that settlers know very definitely the
price at which wild land around them is being held for sale.

When the housewives were asked to break up their general
answers about the meat bill into expenditures for different classes
of meat beefsteak, reastsbeef, smoked and cured meats their answers
were not so certain. They hays pretty definitely in mind the amount
of their meat bills from their paying of them, but the distribution
of this amount between kinds of meat calls for the exercise of judgment
and not memory. '

The housewives' answer.to the question:"In buying meat, do
you Shop between steres?" was divided into three categories, "much"
28, "none 112; "some" 57. The chances are that these answers de-
scribed the buying pretty much as it was. The information asked for
was a kind that could be furnished. The only indefiniteness about
it was in the definition of 'much' and 'some'. -

The question as to receipts from labor off the farm in
a cut-over land survey, in many cases called for totalling in one‘
figure a good many separate amounts, and the settlers simply made a
lump-sum estimate, as is made clear in the following table: '

Labor Off Farm by Cut—over Farmers.

 

 

‘ Frequency Distribution of Closeness Frequency Distribution of
of Estimates, Amounts,
$100’s 57 Less than $ 149. ' 14
50's 20 $150 - $299. 31
25's 5 300 - 449. 21 a
10's 6 450 ~ 599, 11
5's 1 600 - ”749, 5
1's _g_ 750 - 899, 2
Total - 89 900 and over, __§
Total - 89

Fifty-seven of the 89 settlers gave their estimates in hundreds, 20 in
fifties, etc. The frequency distribution for the question "Sales of
-timber products" looks almost exactly like that for "Labor o‘f the
farm". The estimates of amounts spent during the year for groceries
were to the nearest $100 in 28 percent of the cases and to the nearest
$50 in 25 percent of the cases.

86' ' . >ff<4é

Information as to expenditures and receipts and the like Can
in a large majority of eases be obtained with sufficient precision
for the purposes of survey analysis if time enough is taken tp.get
farily complete information so that receipts and expenditures can be
Cheede against each other. The first set of answers to the questions
will very frequently show glaring discrepancies, but if the balancing
can be done on the spot, a number of these discrepancies can be Smoothed
out and explained. This method was followed with fair success even 110
with infonnation as to receipts and expenditures during the first three
years Of settlers' experience in the cut-over country. The statement
of aesets at the time_of settlement, combined with receipts and ex-
penditures and borrowings during the three years following, was made to
appear consistent with the statement of assets at the end of the period.
The experience which farm management men have had with checking opening

“2d CIOSing inventories amply illustrates this point. The difficulty

-i: that this checking_often times takes a long time and becomes annoy-

ing to the person interviewed.

Certain other questions of a definite type did not work out
as satiSfaetorily in the settlers' survey because they were too person—
al in their nature. About 40 percent of the settlers who were asked
the amount of interest payments made on their land contracts or mort-
gages, specified sums to the nearest dollar, but only one-half of the
remainder were willing to admit that they had made no interest payments;
they either evaded the question or indicated by their manner that they
did not wish to answer it, or made an unreasonable statement. A
similar question as to the payments made on the principal of the con-
tracts yielded even less satisfactory results.. '

The question: ”What cash have you on hand or on deposit in
banks at present?", was answered definitely in about 80 percent of the
cases. The rest evaded by saying that they did not know or in some
other way. Enough of the settlers who answered had only a few dollars
on hand so that it was_not difficult to understand the embarrasment of
the others.

Another type of question is that which distinctly calls for
the exercise of judgment. An outstanding e ample of this is the
question as to the value of the farm. Experience has demonstrated
that in some cases, such judgments are dependable and in other cases
not. In the study of land values in Indiana by the Bureau of Agri-
cultural Economies, the actual sales prices were available,in some
cases and the farmers estimated the vsites in other cases. So far
as available tests could demonstrate, the latter were nearly as
satisfabtory as the former, However, when the farmer is asked to
separate the value of the buildings from that of the rest of the farm,
he may go very far astray. In one set of data of this sort, an in—
verse correlation was discovered between value of the land per acre
and of buildings per acre, attributable to the fact that the buildings
had been relatively overvalued on many farms and underValued on others.
This result is very likely to happen in a region where land talues are
declining. The farmer assumes that the building values should be
based on some theory of cost of reproduction now less depreciation,
which of course is entirely erroneous in such a case,

87

The following table showing the results of an inquiry as to
the value of household goods on farms in Northern Minnesota suggests
that the estimates are not very close, and also a very general tend—
ency to underestimate all items of this kind:

Value of Household Goods.

 

 

Frequency of Closeness of Frequency Distribution of
Estimates, Amounts.

$lOO'S 70 less than $99 24
50's ' 48 $100 s 199 53
25's 17 200 — 299 - 38
10’s 7 300 - 399 15
5's ~1 400 — over _l§
E“ . 143

The value of results of questions of this kind depends a
great deal upon the way in which the question is asked. It usually
is not safe to put a mere heading on the schedule, such as, "Value
of farm real estate", and expect to get satisfactory or uniform re—
turns from the different field men. There is usually one definite
way of stating a question which, if presented clearly, and repeated
if necessary, will get the kind of judgment which is wanted. Efither
this must be done in the schedule itself or in the accompanying in—
structions.

As a result of trying to cover too much ground in a survey,
oftentimes questions are included in bald outline which cannot possi-
bly get worth-while information in that form. If the infonnation is
really wanted, it is necessary to go after it more thoroughly and
systematically. This may be illustrated by a question asked in the
land settlement survey, namely, "What live stock should the settler
_ keep during the first few years?" No doubt the carefully considered
judgments of actual settlers on this subject would be worth obtaining,
but at least ten minutes would probably be needed to get them from
the usual settler. First, there should be'separate questions as to
each of the various classes of live stock. The answer will depend
upon many circumstances such as the condition of the land at time of
settlement, resources of the settler, the forage available in the
community, the price of feed, the Ideal market for farm products,
etc. The answers will be worth much more if the reasons for them
can be ascertained also.

In many cases the best practice is to list a number ofx
items which are to be checked off. An inquiry as to vegetables
grown in the farm garden should no doubt take this form. An inquiry
as to sautees of outside income should probably list all the reason—
ably possible sources. This is the only way to prevent omiSSionS.

Inquiries for qualitative data should in most cases list
all of the various answers that are to be expected, and the field
man should stick with the subject long enough to feel certain which

88

one of the answers most nearly fits. The alternative to this is to
put the questiOn in a uniform way and to record exactly the answers
obtained. There are no doubt cases in which this is the preferred
procedure. But ordinarily any classifying Of answers can be done
best by the field man while he has his man in front of him. An ex—
ample of a question illustrating this point is the following: “What
is the attitude of the settler toward the price which he paid for
his land?" The answers obtained were not capable of being classi—
fied Satisfactorily. The inquiry should have been handled in such
a way as to have definitely classified the various settlers as to
whether they thought the price was "too high" "all right", or "too
low“. Any qualificiations which they wished to attach to their
answers could have been included by the field men in addition to

the foregoing and would have been of some value in the office work
later. Another example is an inquiry as to leasing arrangements.
It would be advisable to have the various leasing systems prevail—
ing in the territory definitely understood in advance of the survey
and listed on the schedule and described in the accompanying instruc— ,
tions. If the field men actually found a leasing system different
from any of these or representing modifications of them, they should
of course record these. The information obtained, however, would
be much more accurate if as much as possible of it were reduced to
definite leasing systems in the field.

Another inquiry illustrating this point related to whether
the land company has misrepresented the land to the prospective settler.
The answers obtained would haye been more usable if it had taken the
following form: "Was there misrepresentation? What was it?"

A type of inquiry of this same general sort is that which

_ relates to attitudes. The most satisfactory procedure WhiCh has thus
far been developed with resPect to questions of this kind is as fol-
lows: First a preliminary survey is made of the subject and a list is
made of the answers representing all the different shades of attitude.
The various attitudes are arranged in order from one extreme to the
other. The person interviewed is asked to read the list eat fully
and check the one that fits him best. Where this procedure is not
possible, a substitute for it is asking one general question and
having the field man read two or three of the attitude descriptions
which seem most nearly to fit the answer given in any case in which
the field man is unable to classify it himself. In the land settle—
ment study, an inquiry as to the attitude.of the settler toward his
settlement undertaking gave only partly satisfactory results because
the various field men soon drapped into Special ways of their own of
classifying attitudes. It was possible to Say, however, that a
certain proportion of the settlers were optimistic and the rest were
not.

An example of questions which do not specify definite quali—
tative answers, but obtain replies which are easily placed in a few
classes is the one reported_follqw§ggreSpecting quality of land:

~.'
Rough ------- 5 , Level —————— 65
Rollings ~ ~ — _,_ 40 Flat ------ 24
Slightly rolling -_21 No answer— — - — 3

89

One of course has a right to raise a question as to how
‘1ignificant these replies really were, in view of the fact them
different settlers had dfffcrent 1deas as to what constitutes
"rolling" or "slightly rolling" land. a

This came out even more clearly in an inquiry as to soil
typeS. What many settlers called "loam", others called "sandy".
This difficulty can be obviated in part by the method of having define
ite categories closely defined in the set of instructions and require
ing the field men to followwup their questions more closely.

Many questions asked fail of their purpose because stated
too indefinitely; for example, "Was the land inspected before pur-
chase?" Although this got a definite "Yes" answer in 106 cases and
a definite "No" answer in 9, it did not go very far toward supplying
the information wanted because the term "Inspection" means so many
different things. Likewise, the question "What foods are most im-
portant in your household?" must surely have failed of its purpose
because of the vagueness of the phrase "most important". The ques-
tion: "What supervision does a tenant receive?" is likewrse altogether
too indefinite.

Questions are frequently asked which the persons interviewed
are not really qualified to answer; for example, the question:"How
many acres in your farm are permanently not fit for cultivation?"

Most of them answered "none", when, as a matter of fact, the great
majority of farms had tracts too stony, or sandy, or swamps, to make
profitable cultivation of it possible. Similar questions are those
which relate to the value of different sorts of goods and services on
farms and in homes. The inquiry in a consumers' study, "Are your
purchases influenced by the price?" is probably open to the same ob-
jection. There were 57 who answered "Yes", 106 who answered "No",
and 35 who said "Partly". The situation probably is that few of us
really know except in a vague general way that is meaningless, whether
or not our purchases are influenced by price.

This same objection can usually be raised to most questions
which ask reasons for conduct; for example, "Why did you become a
settler?" There is no difficulty about getting an answer to a
question of this kind, but how correct is it after it is obtained?
Most of us have fabricated a reason for every step in our past records;
but in more cases than not, some other explanation would be a better
one. The question, "Why do you eat meat?" brought forth answers
which illustrate this point very well. The answers, in the order of
their importance, were as follows: "Like it", "Want it", "Need it"
"Habit", "Should have it", "Does not know", "Make a meal", "Everybody
else does", "Health", "Nothing else to eat", "Nourishing". It would
be interesting to put beside these answers a psycholigist's statement
as to why we probably eat meat.

Other questions probably get answers which are not signifi-
cant because they do not fit the situation; For example, the question
"What kind of meat do you prefer"? does not fit the usual situation,
for the reason that most of us prefer no one kind of meat at any and

90

all times. The answer, (beef 137, pork 41, lamb I3, veal 12) does
probably indicate in e very rough way the prevailing preferences of
the population for beef, but there are many occasions when some form
of the other types of meat are preferred,_ '

There are many other types of questions which get answers
which are interesting but whose value ie.questionable in that they
misrepresent.the situation or do not furnish a basis for a usable
answer. The number of persons in a thousand interviewed who used
"Sunmeid" raisins does not_te1l us how many are using them because
they have been advertised; nor whether the advertising campaign has
been profitable, Not remembering Speeific points about advertise-
menfig of meat does not prove that meat advertising has not been

profitable. . . ,- . - *

.‘fi

91

(.) DAIA OI PRIVATE AND PUBLIQ_RECQRDS\.

'An increasingly important source of data is the records of
farmers, private_business firms and cooperatives and departments of
governments. Most readers of this handbook are familiar with the
work which Dr. Stine is promoting in assembling price and related
data from records of farmers, rural newspapers and buyers of farm
Prod.‘ucts.(x The following reports upon projects at Cornell univer—
sity outline the nature of the problem: , '

1. Data from milk dealers, by H. A. Rose:

Professor PearSOn has requested me to write you relap
tive to the method, cost, and other details of obtaining
data from milk dealers. 'I can only give you my experience
in this work, which has been limited to two major studies
completed, one under way and some minor work largely for
Specific information essential to the solution of some
pressing problems such as rate caées, milk shed extension,
etc.

In general, I have found the use of forms for the tab—
ulation of data from dealer's books useless, because no
two dealers maintain the same type of record and it is
cheaper to make the necessary computations and changes in
the office than in.the field. Obviously the aims of the
study must be kept well in mind to avoid transcribing
useless information, but so far I have found blank tabu-
lation paper the most desirable survey form.

Only among the-larger dealers does one find complete .
records extending over any great period of time, with
the exception of milk purchases, which are frequently
kept for many years by even the smaller plants. The
business is so concentrated, however, that records from
two or three dealers~will frequently cover the majority
of the business in a given market.

Ordinarily, dealers are willing'to cooperate in
such studies, although when first approached they gen-
erally fail to see any advantage to themselves.

After results of one study are available, they are
usually eager to supply data for another. The chief
difficulty in many cases is the inadequacy of records

 

(x) See'paper on "Research in Prices of Farm Products" and Dr.
Stine‘s discussion following, Journal of Farm Economics,
Jan. 1928. . ‘

i'7 n r.,_ .
_h . J" .

of the smaller dealers whose aggregate volume is an im- .s'”
.portant part.of the markct~supp1y. Itris cuetOmaryvéor » ‘
milk.dealers~to_cross—check quantities and values in
their accounting‘work,- Forwexample1.purchases of milk
by-plants must'check with utilization.’ The percentage
of fat as shown by the plant test must check with.the
drip test at the city plant. '

The cost of such work is fairly high. For New York
demand study, the cost of travel was approximately $600.
Clerical help'in New Kerk costiabout $150. In addition,
three clerks here worked on the study for about a year.
This time could have been'shortened considerably by the
use of a tabulating machine, which we now have. -

The deliveries of milk by producers is a figure which
is kept by most dealers and we hope to get it for prac—
tically every plant in the New York milk-shed and nearby
territory.' The information on.each plant will consist of
the following items: month, year, dealer, plant, months
Operated, state and county, railroad, freight sane, pounds
of milk, number of dairies, pounds per dairy, ratio of
November to June, test snd‘grade. we already have this
information copied, or it is being copied for us on about
three—fourths of the milk supply, and we started work less
than three weeks ego.-‘

Obviously, the other fourth of the supply will take very
much longer, because most of the companies left operate but
one or two plants.

Data from country milk plants, by C. K. Tucker:

I have given most of my time since August 1925 to the
collection and compiling of data on the costs of operating
38 raw milk plants, 18 pasteurizing plants, 15 bottling
plants, and 10 cream plants in the-New York milkshed for
one year. I visited the administrative offices of the
several milk companies furnishing us this data, and copied
complete cost records for each plant and got whatever other
general information I could upon the operations of the plants.
Then I visited each of the plants studied and by means of
a questionnaire, completed the collection of the data. The
most important information obtained at the plants was the
use of labor and supplies and a diagram of the buildings
with the layout of the equipment.

Data from apple dealers, by Leland Spencer:
I might say that we obtained from each of 15 firms a de—

tailed record concerning each car of apples shipped during
the past three to five years. The nature of the informa_

tion obtained is shown by the enclosed tabulaticnreardr*"\
Each of these cards isuintflndbdrtU“VE217fi¢hQ information
, concerning a single_lo§-of apples in a car. 'Tn‘the ma-
‘jority of cases,~eeveral lots of apples are shipped in
one‘carlii..ei‘several different varities, grades, sizes
and the like. We have not as yet made a count of the
exact number of cars for which we have records, but my
impression is that the number will run up to between
5,000 to 10,000 carloadswfor the 1926-27 season, with~
somewhat fewer cerloads for the previous years. I
would estimate that there is an average of at least
four lots of apples in each car. -

In addition to the information indicated on the tabu-
lation card, we also obtained data concerning the costs
intervening between the shipping point and the first
sale in the city on carloads of apples that were shipped
on consignment or joint account. It will be necessary
for us to use a second card for the tabulation of these
data.' .

l , . . ‘ .

Regarding the uses of such infonnation. I would say that
we plan to determine the effect of variety, grade and size
of fruit, the kind of ppckage, method of sale, place of
sale, accompanying varifiies or products in the same car
and other factors upon the net price to the shipper. we
are particularly interested just now in the kind of package,
as there is a strong tendency. to shift from the barrel to
the bushel basket .in this territory. we intend also to

)determine to what extent the diff rentials according to
:variety, grade, size etc., are re lected back to the
growers.v - ,>

‘I should hays stated that we also obtained data con-
cerning the prices paid by the shippers to the growners.
These latter data are the least satisfactory of all data
Were obtained from the actual records kept by the shippers
and seem to be entirely dependable. The prices paid to
growers are not ??paally recorded in detail and in only a
few cases uas possible to obtain satisfactory records
of this sort.

With regard to the laying out of the'work, we first
visited several shippers in order to ascertain what records
were kept by them, then listed the items of information that
we desired and proceeded to copy on ordinary tabulation
paper the detailed.records concerning each carloéd'shipped.

 

*~\Date of shipment, consigner, and his address, varities, grade,
' size, number of packages, kind of package, gross and net
price, kind of sale.

94

,J 1 - -‘ .. it": ' 4. "3).“. .4 '.."

The copying of the data reQuired the equivalentrtime/of
myself and two assistants for three~nnnths ;a{he tabula,

-tion of the data, punching of. card, etc., wi11 requ.i.re

something like the time of two girls Ifor six months;>plus
the supervision on my own part. ‘ .

"r

Data on marketing of cabbdfllo, by E. G. Misner: ~”

Most Icf the tables are from the becks of one produce
dealer. These were a very complete, carefully kept -set
of records, extending baCk' to 1894. We have' anothe-r
set of such books in the Department giving much data

- for different farm crops, both as to prices paid and

prices received, extending back also to 1890} We hare
had these books in the office for three‘years, but due

'to pressure of other work have been unable to get‘ them
‘worked uP,- I think much useful information about the

marketing of farm crops can be obtained- from the study
of such records. It is a cheap method of collecting
information toIo, because the only expenditure IisI Ifor
.dlerica1Ihelp. ' "

Data Ifrom tax records, by M. Slade Kendricln

It is my belief that here as elsewhere in statistical
studies, the priCe of accuracy is eternal vigilance. This
is particularly true in the case of those public documents
which are assembled by men of little training; May I ill—
ustrate by -speaking of an experience which I had today?

In going through a supervisor's report for a county in this
state, I found a table which purported to give the totaL
compensation received by supervise!) for their services in
1926. One item in this made me suspicious that not all
campensation was included.-”Théreup0n I searched through

-the report and found $3,200 which-had.been left out of the
' table. But errors sometimes-creep into a public report

compiled by a competent statistical staff.‘ For example,

'in one of the tables of general property taxes published
-by a good state tax cemmiSSion, local school taxes are

included for only a few years and then dropped altogether.
There 33 nothing in the heading of this table or the foot— '

~notes attached to it to give one an inkling of the treat—

ment given sshool taxes. There is, though, a significant
difference in the tax totals which make the division be-

'tweeIn the inclusion and exclusion of sohdo1 taxes. ‘This

should make One suspicious.

Personally in spiteof the difficultzves., with public“
records, I believe that they are worthy of use for star
tistical purposes. There are abundant possibilities- of
error in any data and I do not think that public reports
are more liable to error than other statistical material.
Often there is the difficulty that,the Heterial is not

95

quite in the form which the investigation would prefer.
Bit this is true of most statistical material except that
gathered by the survey method.

6. Data from farmers' diaries and Journals, by James E. Boyle:

_We discover and salvage these old Journals and diaries
chiefly by means of an educational campaign put on thru
our fifty-five Farm Bureau Papers and the local and met—
ropolitan press of the state. Twice I have conducted
state-wide contests offering small cash prizes for de-
scriptions of material of this kind which.individuals
happen to possess. When we locate an old document, we
submit three options to the owner:- (1) That he loan us
the material long enough for us to copy it: (2) That he
deposit it in our library subject to call when he wants
it; (3) That he give it to us outright. In this way we
have had several hundred old record books pass thru our
hands and we have perhaps twenty-five or thirty books
which have-been given to us. Rarely_do we actually buy?
material'of this kind._ we did recently pay $25 for a
farmer's diary covering 55 years, daily entries, and
filling five large volumes. Our first aim is to salvage
this material before it is lost or destroyed. It is very
rapidly disappearing forever.

A great many of our records consist of Journals or
ledgers of small country stores. We assume that such
data are almost 100 per cent dependable. Also such
records as farm ‘ages, salaries paid school-teachers
and usually prices received for products sold. How this
material is compiled depends, of course, on the use we
wish to make of it. A recent doctor's thesis on the
history of agriculture in this country, for instance,
was based largely on something like 100 of these old
account books gathered from this one county. The mater-
ial was compiled both topically and chronologically. ThiS‘”
perhaps gives you some idea of the work. I am satisfied
that the six New England States should be combed in order
to salvage and preserve these fast disappearing records.

Using Business Records of Farmers' Elevators.*

This report deals with the use of business records in connec—
tion with a 5-year study of farmers' elevators in the Spring Wheat Area
begun with the 1924—25 season. For the first season, records were ob-
tained from 40 farmers' companies; for the second season, from 61 com-
panies, and for 1925-27, from 96 companies. North Dakota, South Dakota.

 

* By W. J. Kuhrt, Division of Cooperative Marketing, U. S. Department
of Agriculture. _ ‘ '

96

~\Montana and Minnesota have cooperated with the Bureau of-Agricultural
Economics in this-study. About 80 percent of the elevators included
may be considered as operating on sicooperative plan, since'they have
provisions which limit the percentage of dividends payable on capital
stock and permit the payment of patronage dividends if funds are a-
vailable. . . 2

The eleVators studied were selected on the basis of the
completeness of their records, the attitude and interest of their
manager and other officials. and their_geographical location. An
effort was made to obtain a sample stratified according to types of
grain ETOWing. Data on costs of operation, variation in protein
content, and general information were obtained from practically all
of the 96 elevators studies in 1926—27, but satisfactory records of
hedging operations were available and usable at but 46 of them.

- 4 Two men in each state did the field work.in.1926-27. The
forms for recording the data were prepared by the Department of Agri-
culture. In the Spring Wheat Area it has been found expedient to
begin field work about June 1, and an attempt is made to complete the
field work each season before'the new crdp Movement begins. At the
conclusion of the field work some time is spent in copying audit
data. Much of this material is available at the offices of auditing
companies located within the area. Forms have also been prepared for
taking off these data. V

Complete detailed information is obtained covering organiza-
tion and operating practices. The data are planned so as to permit
special study of certain major problems confronting the group as a
whole, such as those of incomes and costs of operation, storage,
hedging, and variation in the protein content and premium values of
wheat and durum.

For analyzing incomes and costs of operation. data are
obtained to cover: (1) volume and sales of each grain and each side~
line handled; (2) expenses of operation, including an estimated a1-
location between the grain and sideline business handled; (3) gross
trading profit or loss from each grain and sideline handled; and
(4) general information with regard to equipment, services rendered
and other related factors. These data permit rather accurate and
comprehensive determination of the costs of operution of each ele—
vator, net income from the grain and side—line business handled,
and costs and income per unit handled.

Storage operations haye been analyzed in detail only for
the 1924-25 season. Records of all storage transactions-were obtained
showing the bushels involved, length of storage period and storage
charges earned, collected and waived.

In the analysis of hedging operations, an attempt has been
made to determine (1) how closely each eleVator actually kept hedged
in each grain throughout the entire season. This necessitated build—
ing up daily "long and short statements for each elevator in each
grain as at the close of each day during the season. From these

97

statements, it has been possible to compute: (1) the losses or gains
from price changes on all grains not hedged Completely, (2) the actual
financial- outcome to the elevator, (3) the probable financial outcome
if the elevator had kept completely hedged, and (4) the probable fin-
ancial outcome if the elevator had used no futures for hedging. It
has also been possible to compute the actual losses or gains of clever
tors from "spreads" between cash and future prices and between futures
used for hedging purposes. All of this information is necessary if a
complete and accurate analysis of the hedging operations by farmers'
elevators is to be made.

In obtaining field and audit data to permit such an analysis,
it has been found necessary,to obtain: (1) amounts of actual grain
purchased each day; (2) amounts of actual grain sold each day; (3)
complete records of future trading transactions including open trades
at beginning and end of each season; (4) market positions at beginning

(and end of each season and much other general information.

Beginning with the 1924-25 season, data have been collected
which permit the following analyses to be made: (1) extent of varia-
tion in protein content between wagon loads delivered by certain farmers,
(2) extent of variation between cars sold, (3) premiums or discounts
paid fer each car sold, (4) computation of the changes in importance of
certain of the important quality factors affecting premiums and prices
paid at terminal markets from time to time throughout the season, and
(5) methods used by elevators reflecting premiums or discounts to
growers. Special attention has been given the last- named feature during
the last two seasons, since it has become one of the most serious prob-
lems with which farmers! elevators are at present confronted.

Costs of obtaining such records vary of course with the number
of records obtained, the quantity of material to be obtained at each
place, the completeness and accessibility of the records, and the sal—
arise as well as experience and efficiency of field men. It has been
found economical to have two men travel together by automobile. Two
experienced men can usually obtain a complete record from an elevator
handling around 200,000 bushels of grain in about two days. If cer—
tain of the time—consuming records, such as, for example, hedging
data, are omitted for any reason, the time may be materially reduced.
A "short" record may be obtained oftentimes in less than a day. The
copying of the audit data necessary for a complete record usually
requires.about two hourS. In copying certain types of records, it
has been found very expedient to have the two,men work together,
one reading off the data while the other copies. This works to
advantage because of the difficulty experienced in keeping one's
place in both account books and research forms at the same time and
without errors.

I _1 ' i '. Table 1.:- Estimated schedule of expense, 1926- 27

 

 

 

 

: Bureau of Agri- : Four State :
v. . , : . cultural : experiment :
; Items of expense : Economics :: stations : T0£al
. Salaries : M .1,1oo.. : $ 2,800 : $ 3,900
Subsistence : . ' 575 _: 1,775 : 2.350
TranSportation : 525 : 900 ' 1.425
Clerical : 523,475 :_ 200 : 3,675
: : :
~Total § : $ 11.550

'$:5;675 : $ 5,675

 

”N

' s H The estimated schedule of expense for the 1926—27 season is

shown in Table 1. ' This.was the first season in.which all four of the
Spring Wheat states participated. 0f the 95 records 45 were complete
and 51 were "short"recorde. :The approximate average cost of complete
records. including office Work, in 1926—271Was'$146, and of the "short" 9

‘recorde $36 ' With a heavier run of grain, the costs would be higher.

One of the most serious of the problems encountered that
many managers of farmers' elevators hesitate to.give out information
concerning certain phases of their business, such, for example, as
their hedging policy or their methods of -meeting competition. This
condition is aggravated if the interviewer_ is a stranger to the manager.
In some cases, the desired information has not been obtained, or state-
ments made have been unreliable. In some this difficulty
has been handled by obtaining the data necessary to check verbal state—
ments made by managers.

Another difficulty is that many of the records kept are not
complete and accurage enough for Some parts of the analysis. It has
been found.difficult to stimulate the interest of.managers enough so
that they will keep the Special records needed. ’Calling a meeting of

-managers and directors at which the plans and methods of the project

have been discussed, has helped somewhat. The interest of the man—
agers has increased as the study has progressed. This is probably
partly due to interest in the preliminary findings and partly to

‘personal contacts and discussions.

Considerable difficulty has been experienced in obtaining
complete audit data. The attitude of public accounts in the area has
been unusually favorable and for the most part their reports furnish
the necessary information. But not s0.the so—called "local" audits,
made by‘local accountants, bookkeepers or committees appointed by
boards of directors. Such audits are seldom complete, are lacking
in uniformity, and are often inaccurate_because 0f inexperience or
lack of proper training in accounting. Whenever possible, elevators
having local audits have been omitted from the study.

The final problem encountered is that in order to obtain
complete and accurate records, elevators must be selected which nodoubt

are above the average of elevators in the area in many phases of their xi
operations. This applies particularly to bookeeping and accounting

methods. But greater efficiency in these activities is often ac-

companied by greater efficiency in other resPects. Thus the study

will probably set up a standard of efficiency which is well above the

average of the entire grouprin the Spring Wheat Area.

Usingggommercial Audits as Sources of Data.*

The primary purpose of an audit is generally to show the
financial condition of a business; 'Hence this part of an audit is
generally the most complete. Experience with audits of Minnesota
business firms (which have been chiefly local marketing enterprises)
shows that the standard classifications indicate the methods of fine
ancing, but give little'information as to the sources of capital. They
frequently give Only scant information as to the method of valuation x
of the assets, and are partiCularlyflacking in detailed infonmation ,
that explains the character of the assets. ;

r

A comparatiVe analysis of balance—sheet information is help—
ful, however, in pointing but the fundamental policies in financial
administration, and sufficient supplementary information may often be
obtained from an audit to explain variations in the distribution of
assets or liabilities among the major groupings of items. Relatively
large figures for "accounts receivable" may be explained, for example,
by supplementary exhibits indicating a large sideline business and a
policy of selling sidelines for credit. Likewise deferred charges
may be explained by a supolementary statement of insurance policies.
Or the relation of the "fixed" to the "current!“assets may be ex-
plained by the nature of the commodities handled as given either in
the weight statement (as of a farmer's elevator) or in the detailed
statement of income.

Such comparative analyses, although they do not explain
financial practises in great detail, are_nevertheless worthwhile in
pointing out important considerations in developing a financial pro-
gram, and they do give a sufficiently fundamental index of financial
policy as to'—provide much information that is useful in extension
programs. (See Minnesota Bulletin 224, Management Problems of
Farmers' Elevators)

The "operating" statement of standard audits gives detailed
information of costs and incomes (both gross and net), the two prin—
cipal tests of the efficiency of a business firm. But like the fin—

,ehcial data mentioned above, they are generally not sufficiently com-

‘ plete to permit anything like an adequate economic analysis of the F
factors affecting the efficiency of operation. The data on total
costs and incomes are accurate, and the items that make up the total
are also generally given. It is therefore generally possible to as—
certain the reasons for variations in costs and incomes to the extent

 

*. H. B. Price, University of Minnesota.

‘\\

-100

\\\‘dflfinrthflyv aIB.made up of component items of outgo and inco
var1ous parts vary between different business units.

\

me and these
eitemized costs

and income Of a farmer's elevator business for a given -year, as presented
in e _commercie1 audit, ‘is—likely to be about as follows: '

I

Qggrating Expense of a Farmers' Elevator

Manager's Salary...... .......... . ...... .
- Director's fees.;...............u.. ....... .
Repairs.......n. ...... . ..... -....,.........3
‘ Rent. ..................;......a....;.g.
Depreciation ‘ . ...... -.....;.....;.....

'mditing& bOOkkaeping...o.......-...-o.-n-

'.P08ta@cnoioa ....... oucw'e-oeoo'I-T-vzoo-i'nnoel-co,-

'StationarY' &print1ng......u vane-oin'ionuno"...

Telephone.- -te1egxaph.......................
Legal feeS.................................
Depreciation, furniture....................
Licenses and bonde..........;.........-....
. Market reports....a......a..;.......;.u..;.
' Income- tax........:. . .....5 ...... ,.........

IDonations and publicity.......;..;....;....i.

Clerk.salary................-..........;...
Extra labor....................a.;.........
Power. light and heat......’....'...............
Taxes......................................
Insurance............ ..... .........;.......
Interest..............;.;.....I............
Miscellaneous.................... ..........

Operating Income of a Farmers' ElevatorL

Wheat-uuni-aq-o.-ulono¢oootone-G:u._..-.‘i‘.oo'-II

FlaonIoOQ-eoIelcs‘o'ooocIncine‘oouooeuoppvaoel“

Rye.....................,..................
oateoIn-ooncqInfutooeuv-b-ceaccoo'nsconce...
Barley...,..;......g..........;........,...
Corn..................................;.;..
Foel.......................................
Flour, feed, & seed.............:..........
Grinding...................................

,InterestS‘and discounts...................._

Mi'scellaneouSCIQOOOlee-looI’OIOOI:OIIIIIUOICD

. $2400.00

188.00
-129.00

-~ _ 55.00

‘ 799.00
-172.00
33.00
45.00
x 172.00
’. 31.00
35.00
10.00
_;-90.00
.309.00
'342.00
1200.00
A 1389.00
(1850.00
‘708.00
744.00'

' 357.00
’ 284.00

$_.217.00¢
..1689.00(
m2&00*

' 11542.00

,245.00
6026.00
1553.00
2097.00
3091.00

459.00
574.00

 

1 Loss‘

- --r - _ '\

\‘

\. 101‘ ~-~ . ‘v. .
.—~_\ .' ' “‘.-.’. ."t7 --J )l‘ W'rh ~
N . \N.
SWr other More will give us a basis for comr-

\ paring the expenses and pointing outwsame of’the reasons for varia—

‘\tions in efficiency. One obstacle in the Way of this is that dif-
‘ferent classifications of expenses.and incomes Are frequently used,
andvoceasienally different kinds of information are included under

. the same items for different‘business units,nwith not enough sup—
‘plementary data to enable the research worker to make the necessary
adjustnwnt of data between items. '

But such information as the foregoing will not permit a full
explanation of the reasons for variation in an item between busi-
ness units. Take the item of labor, for example. One may be rea-
sonably certain that the figures for this expense are comparable,
but one needs to go back of the figures to discover whether it
varies because of differences in wage rates, amounts of labor used,
orwadjustments of labor supply to a seasonal business. An audit
report is generally silent on such subjects.

Even though adequate informatibn for purposes of analyses may
not be included in these audits, what they do contain is so easily
secured as to repay Well the effort.¥

Such data analysed each year may also at small expense indicate ,
the historical changes taking place in the industry or field of busi-
ness. The accompanying summary shows ever a 4-year period the
change in the important tests of efficiency of elevator operation,
namely expense, gross income, net income and volume. That import-
ant test of financial administration, net Worth, is also given.
Grouping these reports by districts would give geographical sig-
nificance to the results. Volume of business varies from year to
year, and with it the per unit cost of operation, the net income
and the net worth. Gross income (or operating profit) tends to
remain fairly constant from year to year, fluctuating much less
than cost of operation, and therefore than income. Not worth seems
to be slole but surely increasing, which indicates an ability to
meet competition successfully, and a conservative financial policy.
Such conclusions amply justify the use of such data in projects
based solely upon them. As to their value as a source of data to
supplement field records, no evidence is needed.

Summary of Expense and Income
of Farmers' Elevators in Minnesota
by Years.
(in cents per buShel)

 

 

Gross Net -

Year Expense Income Incohe Volume Net Worth
1923-24 5.6 6.2 0.6 166,566 $15,868
1924-25 4.3 7.5 ' 3.0 205.800 19,703
1925-25* 4.7 5.8 ; 1.1 223,252 22,985
1926-27* 6.2 7.9 1.5 139,622 24,299

 

*Excludes elevators handling more sidelines than grain.

w

.,'_,
of

w", ‘ ‘ '. _ .
SamPlin§:\'x' \ "Frequently the—busbnecs units includ~
'“' ed in a study of this kind constitute near-
ly‘a cqpplete universe. For cxarple, it
zay'be possible to obtain business records
for as much as three—fourths of the total
volure of business handled in the.area. Certain sorts of generali-
zations can be very safely rade vith.such a large proportion of the
volume of business, but other sorts cannot. For einnple, date upon
daily and seasonal consvrption of milk can be generalized safely on
such a basis, but probably not the data on delivery costs, since the
very firns which were omitted are likely to be following very dif-
ferent practices in this respect.

there a considerable number of business units are involved,
then the principles of sarpling apply in their usual way. The
infiger has already been pointed out that only the better units will
be included because only these have usable records. In wany cases
'here Will be no escape fror this situation, and the only procedure
possible is to wake clear that the conclusions apply only to units
of this grade. If a certain a cunt of information can be obtained
from the rest of them, it will usually be possible to indicate to
what extent the conclusions applv to the whole group. It is advi-
sable in many studies of this kind to select the business units in
a ruch were nearly random ashicn than at present, and obtain sore
sort of information fro“ all of they, :orking it up in corplete
form only for those which are cblc to give it corpletely. The plan
of sectioning the thole revion and covering all the units in cer-
tain portions of it should also be followed in some cases.

There will be more situaliens in rhich the :othod cf stratify-
ing according to sets of conditions as above outlined Till be desir-
able, than when farms are being surveyed. (See the discussion of
this under the Survey Method.)

Accuracy: In working With records of this sort,

no one consideration is more important than

" that of definition of classes. Those tak-

ing off the records must be absolutely

certain as to the practice followed in
classifying various expense and receipt items. As pointed out by
Dr Price in his discussion of grain elevator records, different con-
cerns follow different systems of classification. Norking out a
uniform system of classification into which the records of the va—
rious firms can be made to fit usually proves to be one of the most
difficult parts of such a study. A certain amount of arbitrary
splitting of items often proves to be necessary.

In such work the mistake is often made of grouping the items
into certain large classes when the records are being transcribed
and not recording the details of What haS'been so included. This
frequently destroys the value of the data. The best'practice to
follow in transcribing records of this sort is to bring all the de-
tails back to the research laboratory.

103

_ Mere figures in dollars and cents of receipts and expenditures

~oftentimes have.very little research value. It is the relation—
ships which can be established between these and other factors that
are significant. Some of these other factors may relate to physi—
cal volume of commodities handled, physical facts about layout of
the plant and equipment, details as to organization of the work and
the personnel of the business and distribution of labor among va-
rious activities. Oftentimes much of this sort of information can—
not be obtained with a high degree of accuracy. More effort spent
upon getting such information with reasonable accuracy is often more
to the point than.getting expenses-and receipts accurate to the last
cent.

,1

\ - ..

I - ~13) SUPERVISED'KECORDS .

Because farmers, small marketing agencies and the like rarely keep
financial and operating records which are adequate for the purposes of
research, workers have frequently resorted to the practice of assisting
them in the keeping of such records or even keeping the records in large
part for them.

The most common procedure in this respect is to introduce a stand-
arfltzed system.of records. Regulatory agencies, with power to require
uniform records, have long follOWed this practice. Trade associations are
following this practice increa51ngly. The plan was early accepted in the
field of marketing. In the beginning, hOWeVGr, rarely did enough marketing
units adopt these systems to furnish any considerable body of research _
material. Recently it has met with much more success in this field. Iowa,
for example, has a considerable number of livestock shipping associations
keeping such records; and a group of about fifty creamgrles has just com-
pleted one year of records. Illinois and Ohio have also been successful
in getting uniform systems used by various types of marketing agencies.

No doubt this method of obtaining date for marketing units Will be used a
great deal in the future as more funds become available for research in
this field. ‘

The difficulties encountered are easily imagined. One of the most
important is that a system‘which will satisfy the largest and best managed
units Will not suit the Weakest and poorest managed. Tt is feasible in some
cases to introduce two different systems. Also, generally speaking, the units
which will keep any kind oi record are considerably above the average. The
results may have research value for the last three of the four objectives
named earlier, but surely not.for the first objective, namely, to present
an accurate description of conditions. Another difficulty is that managers

or secretaries are constantly changing.

It is usually necessary to visit all of the units adopting a uniform
system at least at the time of its installation and at the close of the
first year. Additional visits may be needed early in the first year, and
there is likely to be a considerable body of correspondence for a while.

An example of the use of this method in Iowa with household activities
is described by Dr. Elizabeth Hoyt in the Journal of Farm Economics, April
1927, (p.218~20). At that time, 62 families were holping to keep such records.

In the field of iarming operations, the more usual practices along
~the line of supervised records take the form of route studies, the farms
being visited at intervals during the year as well as at the beginning
and end, sometimes as often as once a week. Mr. J. R. Hutson discusses
this in his article just below. The subject is not complete without’;
including the sort of farm financial and production records that Ill-
inois and a-numbenfof other states are trying to get large numbers of

105

their farmers to keep. These records are supervised to the extent of
having them checked over carefully in a central office. No doubt the
major errors are detected in most cases. bentien must also be made of
the poultry and other similar records that are being kept in a number
of states, usually under the inspiration of extension specialists. Tho
research value of these is narrovly restricted to certain objectives,
but nevertheless real for these objectives.

(*)

Route Records of Farming Operations.

Supervised fare records of the route type may be divided roughly
into two groups, (1) these closely supervised and (2) those not co
closely supervised. The kinds of information obtained and forms used,
the frequency of visits in supervising the records, number of records
included in the unit of study, the cost, the kinds of information best
obtained and the uses of which they are best adapted will be dis-
cussed briefly for each group. '

 

Elosoly Supervised 1. Information obtained and forms used:
Records The information usually obtained

by closely supervised records and the
forms used, include the following:

(a) An inventory of the farner's resources including real estate;
equipment, work stock, other liveStock, food, materials and
supplies on hand, the farm.labor supply and notes on the man—
agerial ability of the farm operator. The inventory is filled
out by the supervisor when the record is started, usually from
January 1 to harch l, with the assistance of the farmer coop~
orator. The following details are shewn:

Real Estate-The number of acres of land and its estimated
market value are recorded. All buildings are valued sop-
arately. The water and lighting systems are valued sepa—
rately from the buildings to Which they are attached.
Livestock-Tho name, age, approximate Weight, and estimated
market value of each head of workstock, each dairy cow and
each beef cow is recorded. The other livestock is usually
grouped by claSSes, ages, weight and breeding, and the value
of each group recorded.

Machinery-The number of pieces, size and value of each of
the larger pieces of machinery is recorded. The value of
srall tools, such as carpenter's tools, blacksmith's tools,
etc., is usually recorded by groups.

Food, Materials and Supplies—The amount and farm values of
all feeds, materials and supplies are recorded.

Growing crops-The number of acres of each of the crops for
which labor has been performed the previous year, the amount
of this labor and the amount and value of the seed and for-
tilizer used for each of these crops is recorded separately.
The man labor and horse work is reported by operations and
the times-ever and approximate dates for each operation shown.

 

 

* J. R. Hutsen, Bureau of Agricural EconomicS, U. S. Dept. of Agriculture.

106

Managerial Histories—The managerial history and labor
supply card is filled out at the end of the year by the
supervisor.

l
(b) A clear dayeto-day picture of the work performed on each

(C)

(d)

farm, including the amounts of man labor and horse or trac-
tor power used by operations, with the kind and size of the
implement used, for different crops and kinds of livestock.
The complete time of each regular workman is accounted for.
This information is recorded daily in a labor book by the
farmer-cooperator and checked and taken up by the supervi—
sor at each visit.

The amounts and market value of the different kinds of feed
and pasturage consumed by the different classes of live-
stock including rations and feeding practices. Ration
changes for each class of livestock are recorded when made
by the farmer—cooperator on ration cards or sheets. The
ration reports are checked by the supervisor at each visit.
In determining the rations used, in some cases the concen—
trates and concentrate mixtures are weighed by the farmer;
cooperator and in some cases they are weighed by the super—
visor. In other cases, careful estimates are taken. The
rations of hay and rough feeds are weighed in some cases "
and in other cases the Weights are carefully estimated at
the time of feeding. Usually a monthly statement is filled
out by the supervisor with the help of the farmer—cooperator.
These statements show the amount of each kind of feed used
on each kind of livestock during the month. In addition to
the yearly feed inventory mentioned above, often two other
inventories are taken, one shortly before hay is harvested,
and another shortly before corn is harvested. The ration
reports and feed staterwnts are adjusted on the basis of
these feed inventories and the purchase and sales of feed
during the period.

During the pasture season the supervisor with the
assistance of the farmer—cooperator makes a record of the
pasture changes at each visit. These records show the
field on which each class of livestock is pastured, the
number of days on pasture and the number of days of pastur—
age furnished by each field during the year.

Crop histories including the amounts and costs of seed,
twine and other materials used by different crops, the
amounts and market value of yields of the different crops
and the times—over for the different labor operations with
the size and kind of implement used and the number of acres
covered for the different operations. ‘

A crop history sheet is filled out for each crop, and
when radically different practices are used on different

fields of the same crop, a separate report is made for each

,field. In the latter case, the labor for each field as Well

‘ae the crop histdry data are also reported separately.
These sheets are filled out by the supervisor with the
assistance of the farmerécooperator. The amounts of seed

(e)

(f)

(s)

(h)

(i)

are.kept by someone in the household. The different items

107

and fertilizer used are recorded at the first visit to the
farm after a crop is planted. The number of times-over

for each operation, together with the size and kind of im-
plement used, is recorded at the first viSit after the par-
ticular operation is completed. The yields, likewise, are
recorded at the first visit after the crop is harvested.

An effort is made to Weigh the production of some crops on
some farms each year. Usually in the case of small grain
the bags are counted and a few weighed. In the ease of corn
and hay, a.few wagon boxes or loads are weighed and a record
kept of the total number of loads.

A Weather and soil condition report including the amount of
rain falling from 6 A. M. to 6 P. h. and from 6 Pr M. to

6 A. M. each day and a general statement as to Weather and
soil conditions during the Week. This record is filled out
by the supervisor at the end of each week from notes made by
farmer-cooperators on the labor records.

Livestock histories, including the methods and practices
followed in handling the different classes of livestock,
the calving, feeling or farrowing dates for the different
classes of livestock, the Weights and market-value of ani-
mals when transferred from one class to another or when rad—
ical changes are made in feeding practices, the weight and
cost of the different classes of animals when bought, the
weights and receipts of the different classes of animals
when sold, and explanation of variations in gains and feed
consumption.

Monthly livestock history sheets are filled out by the
supervisor with the assistance of the farmer-cooperator at
the first visit to the farm after the first of each month.

Expenses and receipts, including the amount of all items
that can be reduced to quantities and the particular enter-'7
prise to which all expenditures or receipts belong.

An ordinary cash journal is kept by the farmer—coop-
erater who makes the entries at the end of the day during
which transactions take place. This cash record is checked
by the supervisor at each visit.

The farm layout including a farm map showing the size, shape
and location of all the different fields and location of all
buildings and sources of the water supply.

This map is made by the supervisor and assistants. An
ordinary plane table is used and the acreage is calculated
with the use of a planimeter. When the map is completed it
is checked over with the farmer before the acreages are ac-
cepted.

The source, quantity and value of miscellaneous products
furnished to the home by the different enterprises on the
farm.

Cards showing the amounts and value of these products

i‘:

-\

\

__.«-‘.-_...__

\\“uge~reoorded when thay.a:o*takan into the home rather tiafl
when they are consumed. The munberigcjagge produced 61 -h
day is also usually recorded on these cards. A cardrisjnwr’
vided for each month, which is checked at the end of “6
month by the supervisor. "

\\\' ‘ loé ’- V'_.. .. I .

2. Freguency of visits-In obtaining the imformation indicated
above, each cooperator is visited from 2 to 4 times each month.

3. Unit of studwahe number of records that one field man can
handle is“ generally the unit of study. One man can handle from 20
to 30 of these records. Cooperators for these records usually are
selected within a radius of 15 miles of some central point.

4. Cost-The cost 01 the field Work in supervising from 20 to
SC :—3cords as described above usually ranges from $2000 to $2500
per year. The expenses for travel usually are about $50 per month
and the remainder is salary. In cases in which a man is employed as
superviSor with the view of having him become familiar with the
study and conditions in the section so that he may do more effective
extension work after the research study is completed, the expendi—
tures for the field Work usually will exceed $2500 per year.

5. Kinds of information best obtained—Of the records listed
above, labor, crop history, weather and soil and livestock ration
reports reguire most caref‘ul SLpervision. The other records may
be obtained accurately enough for most uses with less supervision
than is indicated.

 

6. Use for which best adapted-Closely supervised records are
especially useful in showing-tho amounts of seasonal man labor and
horse work used for crop and livestock enterprises and systems of
farming, the amounts of man labor, horse work, materials and cash
used in growing crops and the amounts of man labor, horse work,
feed, materials and cash used in handling live stock, and the re-
sults obtained when different practices are used in growing crops
and feeding livestock. They may be effectively used in connection
with experimental and price data in providing a basis for comparing
enterprises and systems of farming. They are used-in Working out
normal production relations for the different crop and livestock
enterprises, and in outlining systems of farming adapted to typical
sets of resources and conditions and in estimating the returns that
may reasonably be expected from different enterprises and systems.
In this connection the chief function of the records is to localise
the experimental data. ~

 

Because of the detail obtained, closely supervised records are
especially adapted to illustrative uses. They are used to show how
successful farmers have combined different enterprises into systems
of farming that have proved profitable and how the complementary,
supplementary and competing relations have been.taken into account
in Working out these systezns. Such illustrations are extremely use-
ful to farm managorent workers. ‘

They are used to illustrate variations in production inputs

-.’- 4—.

5 i ,. \ w»- W

and practices that have resulted in low inputs . The data made

1 available by the labor, crop history and weatherfreperts for crops

‘ these technological extension uses.

and the ration reports for livestock in particular'are adapted to

1

Other closely supervised records-In some cases, detailed data
as'described above will be obtained for only one crop or livestock
enterprise, and inventories, expenses‘and receipts obtained for the
other enterprises. In such cases one supervisor usually_will handle
from 50 to 75 coeperators and visit each pooperator from 1 to 2
times each month. When detailed enterprise records are obtained
for a class of livestock,_the visits are from 3 to 4 Weeks apart
with additional visits at critical times, such as the farreWing
season for hogs or the calving time for COWS; In the case of crop
enterprises, visits are made every two or three weeks during the ‘
season when work is being done on crops and at the end of the year.
In addition to the uses indicated above, usually enough data are
obtained from enterprise records of this kind to permit an analyt-
ical study of the results obtained when different practices are
followed. ' '

 

Records 323 . Information obtained and forms used: The

Supervised information usually obtained by the loss—closely

C osoly supervised records, and the forms used, are as
follows:

(a) An inventory similar to that described under closely su-
pervised records and obtained in a similar manner.

(b) The amounts and market value of the different kinds of
feed and pasture consumed by the different classes of livestock.
Notes are made by the farmer-cooperatcr as to feeding practices.
Detailed rations are not usually shown. The quantities of feed
used over a period of two or three monthsis usually recorded by the
supervisor at each visit. Inventories are used in checking the
feeds as indicated under closely supervised records.

(c) Livestock history records similar to those described under
closely supervised records are kept. Usually records for two or
three months are filled out by the supervisor at one time.

(d) Crop histories include the amounts and cost of seed, twine,
and other materials used by crops,.and the amounts and market value
of the yields of the different crops. These data are usually ob-
tained by the supervisor at the end of the year.

(6) The expenses and receipts are recorded by the farmer—co—
operator in a manner similar to that described under closely super-
vised records. The records are checked less frequently than in the
case of the closely supervised records.

(f) The source,'quantity, and value of miscellaneous products.;
furnished to the home by different enterprises on the farm. These
data are usually obtained by the supervisor at the end of the year
on forms similar to the monthly forms used for the supervised records. \’

\ \... .

2. frequency of visitsdanh,ooopanflun~‘ v ' I ~\ ’
times during the year. \ ‘ , ' l;s VisrtOd fro: 3 V0‘6f#’-"

3. Unit of study—Ono'field an can handle 125 to 1 '
. . ~ ‘ e 50 ds
Eughbas are described above. This number together with a £330 and”
e y some one in charge of the study from the ff' ' - ‘
,ally-tho unit of study. 0 166 is gener

4. Cost -The cost of the field work in supervisin from 125 Ice
150 records as described above usually ranges from $3800 to $3600
per year. If more records are included, the cost will be proper”
tienally larger. Cooperators for these records are usually so'
lacted'within 5 or 4 counties in the same type of farming area.

5. Kinds of information best obtained-Loosely supervised re-
cords are especially well adapted for getting cash expenses and
receipts and quantities of feeds used by different classes of live—
stock. Laber record data showing the number of times-over for each
operation for crops and ration records for livestock are not easily
obtained by such supervision.

Uses for which best adapted-The records described above are
especially valuable when used in connection with experimental data
in providing a basis for comparing enterprises and systems of farm—
ing They are useful in working out normal production relations
for different enterprises, in outlining Systems of farming adapted
to given sets of resources and conditions, and in estimating the
returns that may reasonably be expected from different systems of
farming. They do not make available detailed data showing the
seasonal amounts of man labor and horse Work used by the different
enterprises. ,

In a complete program of farm management research it Will often
be advisable to use some closely supervised records, some more less
carefully supervised records, and some survey records. Generally
enough closely supervised records should be included to provide a
basis for showing the seasonal distribution of ran labor and horse
work for the principal enterprises under usual conditions in the
section. These may be used to supplement a larger humber of less
closely supervised records showing the feed used for the livestock
enterprises and the cash expenses and receipts from different en-
terprises. These records ray be supplemented by survey data for
still more farms in the area in order to round out the sample and
provide data for some enterprises not coVered in the route records.
Production date obtained from.these different sources when used in
connection with experimental and price data will provide a basis
for comparisons between enterprises and for Working out profitable
systems of farming.

;r..,

A. Q‘ _

"use of the Method; Tho prospects are that research Wprk-
are in the future will resort much more
frequently than inLthe past to various
forms of supervising the keeping of finan-

' cial and statistical records.- Such prac-
tices are expensive, but frequently it is the only way to get re-
cords Which are sufficiently accurate. As research Work becomes
more refined, the tendency will be to resort more and.more to meth-
ods giving more accurate data. Early work in any field very pro-
perly uses microscopic methods,.but the time is reached when the
important problems require a more microscopis examination.

This does not mean that some particular type of supervised
route nmthod'will come to be standardized. It means instead that
practices will be developed which are appropriate to a large
assortment of needs and situations. In some cases, narrow re-‘
stricted subjects will be studied for brief periods over a large
sample. In other cases, similar subjects will be studied over a
longer period with a very small group. There will no doubt be
extensive developments along the line of having the participants
keep records for part of the information needed. Public and
private agencies will in more cases be induced to adopt systems
of records which Will furnish data in the form.wanted, so that the
route feature of.the method will largely disappear.

Sampling: The route method can be conceived as
collecting data to be analyzed by the
methods of either formal or informal sta—
tistics, or by the case nwthod. Most of
the material thus far collected by such

methods has represented too small a group to make the methods of

formal statistics applicable. More than this, the determination
of the units to be included in the group has been so at variance
with the principles of scientific sampling that the statistical
summarizations have only limited validity as such. Analyzed,
hoWever, according to the methods of case prOcedure, they offer
many conclusions of great usefulness. This Will be better under—
stood following a reading of the section on "The Case Method" in

a later section of the handbook.

There is no reason, however, Why samples of sufficient size
and representativeness cannot be handled by various forms of super-
vised records, except that for many types of studies and in many
situations the cost is prohibitive. In due time this will not be
so limiting a circumstance as at present.

Accuracy: While there can be no doubt that a
system of supervised records gives more
accurate data than the survey method, it
is very easy to exaggerate the accuracy
of the data obtained even by this method.

Even though a farm is visited once in two weeks, much of the data

recorded is in the nature of estimates and it is very easy for bias-

to creep into them. The system of recording utilization of time
affects the results. Estimates of quantities of feed consumed re-
flect on the field man's ideas of measures. Other sorts of data

.1112”;

«<~*~~“

“neat Jag recorded With accuracy short of an’ unreasonable imam.
“’3 painful hfl’ort. A good sample of this is the consummim ~°f

forage bxwlivostock on pasture‘w The mere recording of. the number :- ' I

0f days Of 'beisture use leaves— roTOm for a Wide‘rtmge of variatioq._....
The digestible nutrients loonsmned'T'hyrhaavyJooders are likely t0~
have a bitxéfin one direction,» andxthom lay-light feeders a bias in
the Opposite direction. Finally, a statere‘rrt of such thin s as the
value 01' Vegetehles consumed by the farm family, of the va ue of
the labor of childish, of the value of'the labor of the housewife
in making clothes for the children, of theproportion of the super-
vision of the proprietor which is given to the various departments
of the business, is likely to be no more valid~than that obtained
by the field man using the survey method.

a)

”x. ..

.“\

“" ‘1»;Ufi

v

(4) MAIIED quEsmxonfillass_.-

I .
:

1

There are two'types of use§\o(wmailndqugaticnnajres which need to
be carefully distinguished, Those in which the questionnaire is sent to
a list of regular reporters, and those in which it is sent to a fresh list
each time. The first is the method of the Division of Crop and Livestock
Estimates and will be discussed in a later section under the head of "Data
from Reporters". . -

The method of approach to this subject of mailed questionnaires has
been as follows: First, Professor J. D. Pope of the Alabama Polytechnic
Institute sent out an inquiry to all the divisions of agricultural econ—
‘6mics in the United States asking for a report upon their experience with
the use of the questionnaire. Professor Pope's summary of this inquiry
is published as he made it. Second. a comparison was_made of the 380
North Carolina returns: from the Federal questionnaire on farm income,
with the 1115 survey returnsgpreviously mentioned. The returns upon the
cost inquiry of the Bureau of Agricultural Economics were also analyzed
to a limited extent. Finally, Mr. J. A. Becker of the Division of Crop
and Livestock Estimates was asked to summarize his experience with the
form and contents of mailed questionnaires. '

The Use of Questionnaires*

A questionnaire may be defined as a set of questions to be answered
by the informant without the personal aid of an investigator or enumera—
tor. Usually the questionnaire is sent byomail, but it may be distribut—

ed in person; in either case it is filled out by the one supplying the
information. ' .- 4*

An inquiry sent to departments of agricultural economics shows that
the questionnaire is used to a rather limited extent in agricultural ecOn—'
omic research in the state experiment stations. 'Some.of the departments
have not used it at all, and most of them apparently have used it only
occasionally. Very few may be said to have used it extensively.

Proper Those who have used the mailed questionnaire'
Egg agree that it has a place in research _ though a

limited one. This statement of course applies to

first studies, not to the continuing type of re—
search in which reporters are used. In general, it appears to have
been useful mainly in supplementing data obtained in other ways. The
mailed questionnaire is a convenient method of obtaining a limited
amount of information from a large number of persons or {rosin small '
useleétedfgrospswhdch is widely scattered. Sines the questionnaire
is necessarily limited in the scope and amount of information it can
carry, it is obvious that it cannot be considered a satisfactory sub;
gtitute for the more elaborate and comprehensive survey and super;
vised record methods. It is also obvious that with the questionnaire

 

* Professor J. D. Pope.

114

there is no opportunity for explaining personally to each individual
the objectives of the study and the precise information wanted in arm
swer to each question. In Spite of these limitations, however, those
who have used the mailed questionnaire have found it useful in econ»
fmic research for the following purposes:

l

(l) Supplementing a study at certain points.

(2) Providing preliminary information necessary in outlining a
research project or preparing a schedule. , a

(3) Obtaining primary data in certain types of projects in which
it is the chief source o£ information.1 '

(4) Enlarging the background of a problem and filling in the gaps
between areas.

(5) Checking up on changes.
(6) Checking secondary data already available.
(7) Disclosing new problems.' ‘

Questionnaires on most problems have their ideal function as a 73
means of supplementing and giving a geographic generalization with
reference to points on which the use of the interview method has pre—
viously revealed the situation in a more localized and intensive man—
ner. The mailed questionnaire is also a means of reconnoitering for
the sake of getting geographic differences.~

If a questionnaire is sent to a list of farm—
Representativeness ers selected at random, the better educated and
more intelligent types of farmers tend to reply,
so that the returns do not represent a random
sampling or a cross section. Besides, many, perhaps most, of the mail—
ing lists used in sending out questionnaires, such as the mailing lists
of colleges, membership lists of cooperatives, names furnished by
county agents, are composed of selected farmers.‘ Those replying re-
present a still further selected group. The farmers who would reply
to a questionnaire on egg production would likely have larger and more
successful poultry enterprises than those not replying. Replies to
questionnaires sent to bankers, wholesalers or leading business men may
be more representative of the entire group than replies from farmers.
While the returns from questionnaires usually represent a select-
ed group, sometimes a selected group is desired. For example, an in—
quiry dealing with fruit growing may he answered chiefly by the more
sucCessful commercial growars, which may be just what is desired in the
particular inquiry. " ' ‘f" "- ‘
Certain types of questions referring to the community rather than
the individual farm may give a cross section regardless of the type of
farmer replying. For example, data on the age of horses or changes in
the acreage of a crop in a community might be determined satisfactorily
from estimates made by farmers as to the situation on surrounding farms.

o
r

In the few instances reported, those who
Accuracy had checked the accuracy of questionnaires had ~
found the returns were substantially correct.
In a poultry marketing study, the trade pract-
ices reported by questionnaire were found to check
fairly accurately with the data obtained from personal interviews. In
some cases the retuins had corresponded with census data, in others with
data on hand collected in other ways. Returns on an inquiry as to fact—
ors affecting the value of land in cut—over farms gave as good returns
as the individual census schedules for similar farms. Quite frequently,
however, it is apparent that while the data may be fairly accurate as an
aggfigge of the farms or concerns they represent, they are not accurate
‘ for any one return to permit much analysis of variations. With
a large number of returns, with no bias likely to be present, the errors
y be compensating 80 far as averages are concerned; but not accurate -
enough for individual record analysis-or for use in correlation studies.

1-

The number of replies which may be expected

Percentage from a given mail inquiry vary with the type of in—
return dividuals to whom it is addressed and with the na-

ture and number of the questions, from practically

none in extreme cases, to practically 100 percent
in case of,a small highly~selected list, such as extension specialists
or managers of cooperatives. Probably an average percentage return frmn
a general mailing list-of farmers is abOut ten percent. A short inquiry
to bankers or a selected-business group may bring as high as 50 percent
returns or more. The better educated and more intelligent the group ad»
dressed, other things equal, the larger will be the returns. The more
readily understood and the more readily answered the questions, the great~
er will be the returns, other things equal.

" According to several replies to the inquiry, from three—fourths to

7100 percent of.the questionnaires returned to them are usable. Careful
preparation and trying out the form in advance will greatly improve the
usefulness of the replies.

The questionnaire should be brief, usually all
Form on one page including the letter. A formidable ape
' ~ pearance should be avoided. It is perhaps best not
to label it "Questionnaire". The questions should
be numbered for convenience in tabulation.

While no arbitrary rule can be given as to the optimum number of
questions, it may be stated as a general rule that the returns from
most questionnaires could be greatly increased by reducing the length.

The ease with which the questions may be answered, however, is im rt—
ant as well as the number of Questions. VA relatively long list 0

questions each of which could be answered with "yes" or "no" or a check
mark might require much less ttne for'filling out and bring larger re-
turns than a shorter form with questions calling for more thought.

The questions should be short and should con—
Questions tain only one clear~cut idea. The words and phras—
es used should be readily understood by those to
whom_they are addressed. Local usage of words
should be kept in mind. Trade terms should be use

~.

when appropriate and they should be used with accuracy. "

r

all" 1'16 ';-;:.'—_.——'~-'.__ .7 -._

/

tithich is more profitable, the production of apples 0? fleli “f"F3?

"“s-

‘\

Above all, each question.should admit-O§~Only one possible inter—
pretation, leaving no doubt_sithar_as:to-meaning or as to how it is to
be answered. The following is an-ezample-of an ambiguous question;

,
.is was to be expected, the answers to this question were worthless.
Another illustrates the extreme care that must be used in laying 0h?
Questions: "Do you grade eggs for size 7 color? _" View of
the answeres to.the latter part of the question were "bravo" or “Whht?”,

it is also useful when possible to organize and draw up the QUtsbw
ions in such a way that they furnish a check.on each othero

Questions are preferable which are to be answered "yes 0T "no", Cr

,by a single figure, or check mark, or.by checking or underlining one

word of-a listed series, ' When a percentage distributibn is called for,
as what per cent of the money borrowed for crop production is used for
fertilizers? for hired labor? for other purposes? it is well to have
the per cent figures racerded one under the other and-a total of 100
per cent written at the bottom of .the column. This will help the one
replying to check his_own figures.

. Questions of a personal nature, as for example, the amount of cash
on deposit in bank, may be practicable in a personal interview, but are
to be aVOided in a mailed questionnaire. Similarly, questions involv—
ing statements of failures or unfavorable returns from an enterprise are
likely to reduce the percentage of returns. Poultry departments have
have found that reports on egg production and mortality fall off when i
the answers do not make a favorable showing. If the data are thought
by the one questioned to involve a checksup on his tax payments or busi-
ness affairs, he is likely not to reply unless he understands the ob—
ject of the inquiry and the confidential nature of the information.

In some cases heading the page with_the word "confidential" may prove
helpful; Questions involving a great dea1.of_bias should be avoided
unless there is some way of measuring and allowing for it.

A

. fMost of those who Supplied data on their use
QQEE - of the questionnaires have used the frank. Such
- studies have usually been conducted in cooper-
ation with the United States Department of Agri—
culture. The use of the frank of course reduces
the cost considerably. If four cents postage is needed for each
questionnaire mailed out, including a stamped envelope for reply,
and a ten per cent return is received,.the cost of postage alone per
questionnaire returned will.be forty cents. Such a cost seems ex—
cessive when‘it is remembered that in some states the federal agrie "
cultural census has been taken by local workers at less than that
cost per schedule. The estimates of the costs of questionnaires‘varied
from 1 to about 5 cents per questionnaire sent out where the frank was
used” - ‘ , » H

A

,The sources of mailing lists for farmers are

Mailing . .' usually.the college mailing lists, names supplied
Lists ' .by the county~agents, and cooperative membership

lists. Names of wholesalers..merchants and other: -
business groups may be obtained from secretaries of trade associations.,,

117

:Names of consumers and dealers are often obtained from city and tele—
phone.directories.

It is important to plan the questionnaire care—
ggneral fully. Its objectives should be clearly outlined.
Its relation to larger problems should be understood.
The general make—up of the form should be tested in
, - advance. The methods of analyzing, tabulating and
digesting the data to be obtained should be carefully worked out in ad»
vance. ,

An excellent plan which is used by some is to try out the prelim,
inary draft of the questionnaire in a small way before sending it out
in final form. If 2000 forms are to be sent out, 100 might be sent out
in advance, the returns from these examined carefully to see that the
information sought is being supplied and that no questions are being
misunderstood. A trial tabulation can then be made of these few re— .
turns as a check on the readiness with which the results will lend them—
selves to tabulation. The form is then revised if necessarykand it us—
ually is. Such a procedure usually results in making the returns from
the inquiry much more fruitful than they would otherwise be.

Experience of the Division of Crop and Liv
Form of stock Estimates in securing information from farm—
Questionnaire‘ are by use of schedules Zrnt and returned through
the mails has indicated the unquestioned superior—
ity of the printed schedule over multigraphod,
mimeographed or otherwise duplicated schedules. The printed schedule
has a distinctly neater appearance and gives an impression of brevity,
and can be written upon either-in ink or in pencil. The percentage
of returns to printed schedules has been uniformly higher than to dupli—
cated schedules.

Some experiments have been made by the Division with respect to the
variation in the percentage of returns to different colors of paper.
There is apparently no great difference in the return to a regular in—
quiry by regular reporters, but in intermittent inquiries and in inquir—
ies to new reporters, the color of the paper has some effect. The re—
gular inquiries sent out from the Division office in Washington are al—
ways printed on white paper and those sent out from the field offices
are ordinarily printed on green paper. The "pulling power" of these two
colors is not greatly different. If these be taken as standards, it is
found that light pink is more effective with new reporters and intermit—
tent reporters, and light yellow slightly better than light pink. How.~
ever, the constant repetition of light pink or light yellow to intermit—
tent reporters apparently results in a considerable loss of the "pulling
power", so that alternation of these colors gives better results than
the use of either color continuously. ”'

Where the inquiry relates.to a particular economic condition or sit-
uation, it is often worth while to include a brief statement of the sit—

 

* By Joseph A Becker, Division of Crop and Livestock Estimates.

118

nation up to the time of the inquiry. For example, in the special in—
quiry made to crop reporters concerning the number of horses of Varlons
ages on crop reporters' farms, a statement was included to the effect
that "current information, and the trends of horse numbers on farms in
the United States, indicate that the number of colts at present being
produced on farms of the United States is greatly below replacement re—
quirements". If the schedule is sent to a new reporter, with the intent—
ion of asking regular cooperation, it has bzen found effective to eccen-
pany the first schedule by a more detailed letter concerning the service,
with the request that he return a blank on which he agrees to make re—
ular reports whenever he is in position to supply the information re—
quested. This agreement to cooperate, which the farmer is asked to sign,
Chbuld not be made too binding. This agreement should read "I will at-
tempt to" or " I will try to comply with the request" etc; If made too
binding, or in the form of an absolute promise, he may talk at signing it.

The inquiry should be brief, or give an appearance of brevity, par—
ticularly tO'new reporters or to intermittent reporters. The regular in—
quiries which are sent out each month from the Division of Crop and Live—
stock Estimates in some cases contain as many as fifty questions, but the
horizontal type of schedule utilized by the Division gives an appearance
of brevity.

The experience of the Division indicates that it is impossible to
secure accurate answers to questions which a:e too closely defined. The
reporter becomes involved in the phraseolcgy of a closely defined quest—
ion and often misinterprets the question by placing emphasis upon one
particular phrase rather than upon the question as a whole. For example,
if information is desired concerning the number of milk cows, and too
exact a definition is attempted, although this definition may coincide
with the gfihmer's definition of a milk cow, it may differ entirely from
the reporter's definition of a milk cow, and the answer may or may not
be given in terms of the framer's definition. Our experience has indi—r
cated that it is much easier to interpret the results of an inquiry

expressed in general terms, making adjustments from collatexul data,
than to attempt to pin the farmer down to the farmer's concept.

If the farmer finds the first question involved or difficult, he
is inclined to stop immediately and not make a return. Acreage sched—
ules of the Division, for example, invariably ask the fanner to report
the total acres of all land in his fann first, because of all acreages

on his farm the acreage in the whole farm is the easiest for him to de—
termine.

_ Within the limitations of asking easy questions first, the most
important question should be asked nest, since our experience indicates
that the first questions are ordinarily answered most accurately and
more thoughtfully. If the most important crops in a given State are

corn, wheat, and oats, these should be asked next to the total land in
farm question.

. .Similar questions should be rou e to ether ' '
editing schedules indicates that fiherg €38 grder 5r 8g: giggiigficgoég

pels the reporters to shift their line of thought from numbers to prices
and back again to numbers, they become confused in some cases report
prices where numbers were asked, and vice versa.

\

119

Each question should be complete in itself, and no qualifications
on the questions implied from a previously stated group heading. If a

repetition of the imnlied heading_results in an involved question, the
question should be simplified. This is one of the commonest faults in
schedule preparation against which even experienced schedule makers
have constantly to guard themselves. An example of such a schedule is
shown below: - .

1. Acres of all land in farm.

WINTER WHEAT
2. Acres planted on your farm last fall.
37 Acres harvested on your farm this year.
4. Acres planted on your farm this fall.

To the schedule maker it is obvious that questions 2, 3, and 4 all re—
late to winter wheat, but the results of the inquiry invariably show a
considerable percentage of farmers who report question 2 accurately re—
lating to winter wheat, and questions 3 and 4 inaccurately by report—
ing acreage in all croos. In this particular inquiry it is entirely
pflssible, because of the nature of the questions to edit and discard
those schedules whereon questions 3 and 4 are incorrectly answered,

but many schedules similarly boxaheaded do not permit of such editing
and the results correspondingly contain an unknown amount of inaccur—
acy. ' '

While it is essential that as few questions be asked as possible,‘
in some cases where there is a possibility of a misinterpretation it is
necessary to inelude an inquiry on subject matter Concerning which no
information is deSired, in_order that the_question concerning which in-
formation is desired be answered properly. An example of such a quest—
ion is found in the pig surveys, where it is desired to secure a com—
parison of the number of sows farrowing on June 1 this year and on June
1 last year, and also the average number of pigs per litter this year,
the number of pigs per litter last year having been already determined
by this same inquiry a year ago. The logical order for the three quest—
ions is as follows: ‘

1. Number of sows which farrowed on your farm between December 1,
1927 and June.l, 1928.

2. Totfil number of pigs saved on your farm between December 1,
1927, and June 1, 1928."

3. Number-of sows which farrowed on your farm between December 1,
1926 and June 1, 1927.

Because of the juxtaposition of the inquiries, it was found that a con—
siderable inaccuracy was.involved because some farmers in answering
question 3 gave not the number of sows which farrowed but the number of
pigs saved. AccOrdingly, it has been found necessary to include a fuurth
~ questiOn on the number of.pigs saved "Number of pigs saved from all litt-
ers farrowed on this farm between December 1, 1926, and June 1, 1927"f“

120

Because of the selectivity of sampling, which is discussed else—
where in this report it has been found expedient to request certain
judgment replies to inquiries which themselves throw some light upon
the skewness of the sample. For example, a farmer reporting the pro—
duction of apples on his farm this season and last season may show 6
some increase in production this year over last year. A Judgment in—
quiry as to production in his locality compared to last year might
show a material decrease. The apparent disagreement between the two
series might lead the farmer to volunteer‘the information that his or—
chard is made up largely of Baldwins which had an off—season last year
and an on—season this year.

Many of the inquiries of the Division are made regularly each
month during the growing season and year after: year. Experience in
editing returns to inQuiries have proved the value of using identical
phraseology in successive inquiries in order that comparability may not
be destroyed. EVen the use of bold type for setting out the important
part of the question is carefully guarded against change.v Experience
has shown that the answers to the question "Number of Pigs Saved" are
not exactly comparable to the answers to the question "Pigs Saved". or

A Judgment inquiry concerning the proportion of the cotton crop
ginned to a specified date has been carried in this order on the echo—-
dules sent to one list: "What percentage of the total cotton crop in
your locality has been ginne.d to December 1?" "What percentage of the
total cotton crop in your locality has been picked by December-l T";
and on another schedule in reverse order. The results of the two in—

Quiries vary by a number of percent, and the variation tends to be con—
stant.

In its work of following up the schedules sent to a reporter, the
Division has found the most effective follow—up method to consist in
sending a duplicate schedule with "Second Request“ stamped directly 3—
cross the face of the schedule in the upper left~hand corner. In some
cases the returns to such a follow—up have actually exceeded the orig-
inal return. 'The. use of a special letter to accompany the second re—
quest has not proved nearly as effective. The most effective method
of securing faithful performance on the part of regular crop reporters
has been a mere slip of paper on which is printed in large type "WE
MISSED YOUR REPORT LAST MONTH". The appeal of this brief phrase has
exceeded in effectiveness all circulars, form letters, or personally ‘
signed letters, so far devised by the Division in maintaining the co~
operation of its regular reporters.

Timeliness of the receipt of the report-by the reporters has been

" found to be extremely important in securing returns. The most effect—
ive results are secured when the report is- placed in the farmer' 3 hands
about two days before the date to which the report is presumed to re—
late. A report of concerning the condition of crops on July 1 is more
largely returned if received by the farmer about the 27th or 28th of‘
June than if actually received on the first of June., There appears to
be a disposition on the part of farmers, and this undoubtedly is a.trait
of humanr-beings in general, to out off making a r‘eply immediately upon
receipt of an inquiry. If, however, the reply is put off for more than -
tWo or three days, the tendency is to .forget it entirely. Returns to J

questionnaire which bears no date on its face has also been found to'
give indifferent returns. It is, therefore, apparent that all the re-
ports calling for information as to current conditions should be dated,
and a statement included as to when the reports should be filled in and
returned.

The general practice followed in the Division in formulating a new
schedule is to put down all the questions which seem essential to the
inquiry and then begin a process of elimination to reduce the number of
inquiries and the number of words used in each specific inquiry to a
minimum. If there is some doubt that the questions asked will be under~
stood and answered accurately by farmer reporters, it has been found
very fruitful of good results to send out a trial inquiry several months
before the contemplated date of the fu11.inquiry to various typical sect—
ions of the United States. A careful analysis and Close study of the in—
dividual reports is'then made of these trial returns. Repeated trials of
this nature have indicated that misinterpretations vary between sections
of the country due to variation in farm practice in different parts of
the country and also due to slight variation in the local use of certain
phraseology. Fbr example, "pigs saved" in Iowa apparently means the num—
ber of pigs which were raised to weaning age, but ”pigs-saved" along the
fringe of the Corn Belt means pigs which the farmer raised and did not
sell to his neighbors for fattening.

It is generally reocgnized that Questionnaires

Sampling obtain returns from a select list of correspondents.
The frequency distribution of_sizes of farms given

in Table l in the section on the survey method pro-
bably gives as good an illustration of this as one

could want. The questionnaire farms averaged 157_acres as compared with
128 for the survey and 65 for the census.' The two groups of smallest—

sized farms were almost entirely left out of the questionnaire returns.

It is usually assumed that because of this the data obtained from
questionnaires has little authentic1ty, Such a statement presumes that
a study can have only one objective, namely, to furnish an accurate de—
scription of Conditions. It has been made clear in an earlier discus~
sion that more description is a low order of research, If the question—
naire returns could be depended upon as accurate, they undoubtedly would
lend themselves to a great deal of research in which the objective is
the analyzing of certain sorts of relationships, which may be nearly as
authentically revealed in a group of the larger and better farms as in a
well distributed sample. Another objective of research is to measure
change, and it is probable that this is represented more nearly correct—
ly by a comparison of the.same sample at two periods, even though it may
be biased, than are the_absolute conditions at any given time. However,
one cannot be sure that the bias does not extend in some degree to
changes and relationships; hence one is always more or less on uncertain
ground with a biased sample.

In the list of uses of the questionnaire method outlined by Profess—
or Pope are a number which do not require a representative sample even to
the extent that the two just mentioned do. An inquiry directed to discov-
:ering what leasing systems maintain in a certain territory, or what land
clearing practices are followed in.a timbered region, or what practicer§7

.r

A, . > -_ fl 1. v "‘ _ .
V; V . . \ KN.— W
.- " . . - - .. . . _,’

~ . ~ 1

are followed by the trade in handiing certain commodities, does not re~«‘*-
.quire a highly representative sample. \It~islnon geocming more and more

.realized that various sorts of data must be kept -cp—to~date. Increas-

ing use is therefore likely to be made of questionnaires to check up or
rh£.ngcs cf various kinds in the details of farm practices, in varieties

J3? crops planted, in dairy cow rations, in croo rotations, and the 11;~

Mr; S. W. Mendum of the Bureau of Agriculture:
3_X, ' - Economics has made a comparison of the summaries o‘

101 fa.rm account books for 1926 in the state of

with 493 questionnaire returns from Ohio farmers +>

' that year. -Unfortunately, the samples are so m ch

different , the book records being mostly confined to a few counties, ‘hat
no significant comparison can be made. The questionnaire returns show 3d.
a wider range than -iid the book records, 8 per cent of- them being belor

AC

(3

L'

‘zero, and 4- per cent of them above $5 000; Mr. Mendum speaks of the

difficulty over the inventory returns in the mailed questionnaires. He
-says: "Few of the corresponding inventories for the same farms, .which

,should be identical, are even approximately the same, though the average

works out someWhat closer than the average of all farms in the state sam-
ples. We have accepted increases each year on the theory that farmers
have really been adding to inventory yearly since the depression years
through better crops, imorovement in prices and expenditures for equip—
ment; but this cannot go on indefinitely without a more substantial veri—
fication." The book records showed an increase in inventory from 1926
amounting to $253 compaied with $162 for the mailed schedules; hence, the
Ohio book records threw no light on this problem.

, Another-item in doubt in the mail inquiry is "miscellaneous expenses
not otherwise reported above", which is too often zero or nominal. For
.the mailed questionnaires the miscellaneous expenses averag-ed $186, or
16.6 per cent of the total expenses, compared with $424, or 27 per cent
for the book records.

Mr. Mendum's general conclusion was that this comparison-did not
change his general opinion "that the showings of the mail schedules are
reasonable and conservative, and an undetermined degree higher than a
census average."

Table 1 compares the returns on the survey and questionnaire farms
in North Carolina. Although the questionnaire farms are two and a half
times as large in acres,.both their receipts and expenditures were a.lit—
tle less. Evidence later nresented indicates that the tax figures are
nearly accurate in‘the questionnaire returns. Obviously, therefore,'the

. questionnaire farms are nowhere nearly as much larger than the survey

farms as their acres would indicate. Even allowing for this, however,
the questionnaire averages seem to be too low. The discrepancy is most
notable in the two miscellaneous items. '

4
.1

123

Table 1.— Comparison of Questionnaire and Survey Returns,
North Carolina, 1927.

 

 

 

Survey Questionnaire
IBeceipts SuIVey
CropS.....'......‘..................$1,339 ;.'.... ....... $1.105
Livestock & livestock products.... 234 ................ 378
Miscellaneous...g.....' ..... ‘ ....... 101 ... ........... .. 26
Total..f.....,.$1,674 . ............... $1,509
.EgpenSes
Livestock bought............‘...... 33 78
Feed-o‘eu-ca-V...udu.n...‘...Vo-oaoa-‘. 71 ou.-v-o.‘..’..--on 82
Fertilizer... ...... ...;.L..; ..... '. 208 .. .............. 188
Machinery.; ..... r; ..... 1....; ..... ‘. 20 ................. 47
Improvements.....;..... .......... , 83 °--a ............. 102
Taxes. ..... 2 ........ ....LL...1....; 97 ....,............ 115
Other Expenses.. ...... .1...{ ..... 3, 194 .. .............. . 136
Labor..... ....... ' ........... u....L 376 ..... ......... .... 296
Total....i.....‘_.$1,o'7‘7 ......... .. $1,048
Value of farmer's and femilyi ‘
- 1abor.._ ..... '- $ 912 ........... $384
Size of farms, acres .......£....i 65 ...... ..... V..... 157

_ After the survey had been made; it was discovered that seven of
the 1115 farmers had sent in'a return on the mailed questionnaire.
Table 2 puts six of these returns-in deadly parallels. The seventh
one is omitted because the farmer reported only 60 of his 260 acres_
in his questionnaire return. The sixth one listed was discarded in
the editing of the questionnaire returns because of the low valuat—
ions on;the real estate.4 The survey disclosed that all that happen—
ed was that a cipher was omitted. :The tax figures are exactly the
same in only one case, but nearly the same in two others. The esti~
mates as to the value of_the food furnished by the farm and consum—
ed'by the farm family are widely apart. ~Tota1 receipts and total
expenditures are quite far apart in most cases and the individual i—
tems are even farther apart. .Tho real estate valuations are widely
apart in some cases. It-is difficult to believe that this is not
an exceptional set of parallel figures. They are included because
they are suggestive rather than conclusive.

9!

 

 

 

 

 

 

 

1“_ ‘. ~-\‘
124
FahTe 11.— Comparison of Survey and Questionnaire Return on
Six North Carolina Farms, 1927
'"—"’-_‘_"1-—"—'_.f“_ .1 ; B : C - : D = E :_u _;5 _____
Vin§t;r-: 11—: Sur—: Q1— :Sur—:Qu~ : Sur—: Qu- :Sur-:Qu— .So.z- : Qu
;_§9y : £33.: 19v : est. :vey :est. : vey : est.:vey :est :19Y_;_§St
_; E :3.~::: . . . . = : : .
Te~ul.1l. ....;$2291: $9700‘32O34: $2120: $154: $250:$5419: $2867: $258: $voc-9§7*~$500
:99: ........ .. : 78f8: 9400: 1978: 2010: 10: 50: 5311: 2715: 258: 600 ’ T ——
f%-:z.k:..::....: 50: 200: 30: 40: 56: 100: 45: 62: ~— ——. 1 —-
" Ernaucts...: 113: 100:., 26: 40: 88: 100: 63: 90: ——: 100: —-
"igjgiganeo1s... : __, -1- _- __ __‘ -_' _-; -- __: __ _— ~—
Larn Exgenses : ' : ,. : : : : : ' : :
Tobal.........:$6154: 5234: 2981: 1863: 348: 207: 3392: 1287: 276: 410: 560: 500
,1c;r...... ..... : 2937: 2150: .925: 896: 15: .25: 1485: 150: 64: 150: 80: ——
.1vestock bought..: 100: 225: .20: 20: ~—: --4 10: 'f-:' ——' -- 6 ~—
1934 ” : 205: 115: 96: 75: 17: ~10: 87: ‘10: 40: 20: 31: ~—
1Erbilizer " : 863: 925: .569: 362: 33:» 30:.1050:‘ 650: 60: 60: 130: ~—
“Fed " . : 39: 5: 28: 8: 10:' 20: 48: --: 2: -— 21: ——
Taxc s ............. : 421: 545: 72: 72: 56: 72: 187: 190: ’77: 75: 165: ~—
NEW Machinery..... 20: 120: 111: 20: #4: 15: 55: 55: ~-: -—: -— —~
N .w Buildings..... 225: 600: 425: 200: 60: 25: 150: 200: ——: ~—: -— ~—
xterest ......... . 582: 425: 315: 360: -~: ~—: 30: -—: ~-: 105: ~-
;;,re11an ou.s..... 762: 125: 420: 50: 57: 10: 292: 32: 33: --: 127: 400
‘liJnLory : ‘ ..:. 7 - - ; : . : . . .

0*ops. ........... .
1111788130015. . u u I
Machinery.. ..... ..

Supplies.........
Real Estate......

End:
”Total

Livestock.........
Machinery.. ..... ..
Supplies.... ..... .
Food from farm....
Farmer 's and
family labor...
égpes..;, ......... .

:eginning, Total:

Crops.... ........ .

In I. on - o. to co a. on

2959: 6700:
: 1200:
944: 16002
1800: 3500:
215. 400'

28658: 28500:

2991: 6425:
: 1100:
906: 1800:
1700: 3000:
385: 525:

281: 600:

150: 1200:

253

243:

650:

125:
400:
125:
8000:

575:

‘0

175:
400:

150

dc .- uc 0-

200:

100:

975:
768:
125:

82:

3800:4000:10000:

1800:
500:
800:
500:

1072:1900:

776:
125:
171:

400:
1000:
500:

_-:

c. u. I. .0 sq .-

H
U
m

1534:
440:
775:
319:

1397:

<o¢o-¢Q¢o.¢.-

'184

823:
300:
450:

75:

1000:

850:
300:
450:
100:

—_

I
|

800

a. .- 4. o. a. c. In

E

656:
193:
160:
103:

1100:
400:
300:
100:

971: 400

——: 200
240: 400
300: 100
131:

3505: 4000: 7025: 7800

500:

. 300:
. 100:

400:

300:

124:

——: 650
-—: 150
250: 400
300: 100
156:

311: 200

300: 400

270: 270

 

 

 

,125,//"r

One outstanding deature of the-questionnaire returns is/their
tendency to round-numbers in many items. ;The table following shows
that in the matter.of tax-returns, there is little difference in this
respect.‘ Fertilizer-purchases are much more largely in round numbers,
and estimatei value of food obtained from the farm are largely in
round hundreds: v -

Frequencies of Number of Returns in Round Numbers,
' North Carolina Data

 

$ Questionnaire Survey
100's . ........................ 23 ..... ............ 22
5 50's .. .. ..................... .. 20 ,.. .......... .... 11
25's ...q. .................... a..; 25 ... .............. 22
10's .... .............. ; ......... L 68...;... .......... . 67
.5's ..... .......... .... ......... a. 59......... ..... .... 41
l's'....,, ............ x. ........ . 185 ..' .............. . 217

380. 380*

 

* Selected at random from.surveyrreturns..

 

 

.Food Value

 

 

 

 

- -: Labor
$lQO's ....., ......... , ..... . 231 ,z $lOD‘s .. ...... ...... 85
.50's .....,..p......,..u... 62 .; 50's ... .. ....... 66
25'54.....,..,; ........ a... 28 ;: 25's-.... .......... 45
lO'S'onnougo.: ....... ...-3... 18 : 10.5. ............. 57
5'5 .... ... .......... a... 9 . 5's .. .... ...... . 15
1's ... ................... 12 : l's-........ ..... . 20
. Feed Bought . -: Fertilizer

$1oors.....,...u......l.n. 42; : $100's...... ........ 81
50's: .....,... ......... y... 47 1: 50's . ............. 57
25's ....., ............ t... 35 _: 25's .. ...... .. ... 42
10's .....,..t.. ...... ..... 63 -: 10's .. ........... . 78
5's- .....,..... ........... 16 .: 5's ... ........... 33
1's~....;......... ..... a... 20 ;: 1's ....... ...... . 57

A consumerfls inquiry_sent Out_in Washington asked the specific
question: What kind or brand of butter do you use? A total of 209
answers were obtained which named a definite kind or brand. This
indicates that a simple question clearly stated calling for facts that
the informant really has, will get1reeults even in a mailed question-
naire. The same informants even tOok time to list an average of over
two brands each in responSe to the request: "Please list all the dif—
ferent brands of butter you can think of." The Question "How long
have you used this brand?" also gave a clearscut response. "Why do
you use this particular brand?" was not a call for facts, and the
answers showed it, although no doubt they had considerable signifi—
cance. -

'125.1.31.
v More than a tendency to round numbers is involve: in the_an~nma!r*fibnlamed
21.10w1ng to the questions on costs of production\of corn.anfi..cotton’1n 1926. (The
:?6 cases are from Ohio Indiana Illinois, Wiscongin and Michigan: the COtton COStS
rom Georgia, Jorth Carolina and South Carblina). 10.5.5 obvious from the wide ranges
'a: many of +he‘ correspondents have a very faint notion as to the corre(.t answer in
'51r 0630, 1fhho e is (n0. It is doubtful if inquiries by QUGStlnhnal’es should
I»? 00 01Tec’ced a: m;1tters ccnverning which the correspondent cannot paisibly have
"53:31 s;:_~:=_:_r_-3:j___ 116161;. _a 1mm. _'._.: v- -
'uw :‘231 fo~*-0 “nfitcw::0“n- :Corn-WEar: Corn :Corn~ M01 :Cntton

 

31‘1V611>“B”Gparing:0rop16nd:Yield :& Tear on:0verhead :Cas 0 Rent: 031:1: Per Acre
3 Acre :0000nd :Sells‘ :Fer ' :Implen'ts: Cost Per Per :1on 10rt33%U1't°
___ :Fq; Seed :Fcr :Acre :Eer..Acre :Acre :Acve‘ " :Acze :Market
‘151 HO : 00st Nozfirmce No: Bu. No: Cost No :Cost :Nor'nent Nd: Cost No: COM NO
.50 2 :$ .50 1: $20.11 1'51:31-10..6:$.10.~4:-$2.5o 4:6 .20 31$. .10 6
.60 1: .80 1: 50 '1: -20. 1,: .20.. 8: .20~ 5: 5.00' 6: .50 ,6: .2'0' 4
.20 2 : 1.00 2: 35 1: 25 3:. .25.: 8: .25 -'6: 3.50 "1:' ./0 _'1;...25 10
1.00 :3 :1 .50 1: 4o. 3: 28 1.: .30-2: .30 2: 4.00"'3: ‘.80 2: .30 1
1.25 l : 2.00 4: 45 1: 30 9: .40. 4: .40 : 4.50: ;2: 1.00 16: .40 :2 '
1,50 7 : 2. 50 1: 50 4:32 1: .42 1: .50 17: 5.00 18: 1.10 1: .50 15
1.60 1 : 2.70 1: 60 3:35 5: .50 ll: .60 1: 6.00 17: 1.20 1: .60 3
1.70 1 : 2.83 l: 70 1:37 1: .80 5: .80 2: 6.50 1: 1.25 1: .80 . 3
1.80 3:: 3.00 -11: 75 6:40- 22: 1.00 26: .90 1: 7.00 9: 1.50 8: 1.00 20
3.00 11 : 3.40 l: 80 8:42. 3: 1.20 1:1. 00 20: 7.50".2: 1.70... :.l.25 3
2.25 4 : 3.50 4: 85 1:45 6:-1.40 1:1.11j 1: 8.00 15:11.80. .2: 2.50 5
2.50 6 : 3.80 2: 90 4:50 24: 1550 6: 1.50, 3: 9.00 1.2: 2.00 ”1.2: 3.00 1
3.00 16 : 4.00 12: 100* 21:52 2: 1.80 2: 1.80‘ 1:10f00 15; 2.40 .1:.4.00 1
3.25 1 : 4.25 3: 105 1:55 34-2.00 8:2.00 ‘ 3:12.00 _3: 2.50.. 8: 5.00 5
3.40 1:: 4.50 1: 110 1:60 9:-2:40' 1: 2.50 ' 1:15.00 1: 2.80 2: 9.00 1
3.50 1 : 4.60 1: 115 1:65 5: 2.60 1:3.00 8:20.00 2: 3.00 19:
4.00 7 : 4.65 1: 125 5:70 2: 2.80 1:4.00 3:20.20 1: 3.25 .1:
4.50 1 : 5.00 17: 140 2:75 '1: 3.00 1:5.00 _ 3:25.00. .1: 3.50 3:
.60 1 : 5.25 2: 150' 15:80. .1: 4.50 lzno ' : ' : 3.80 ..1:
5.00 9 : 5.50 3: 175 1:90 1: 5.00 2:reeord 2: 2: 3.92 . 1:
5.85 l : 5.80 1: 200 9: : .. : : : 4.00. 11:
6.00 4 : 6.00 13: 250 . 1: : : : :_ 4.25 1:
6. 25 1:: 7.00 5: 500 2: :. : ': 1:41.40. 1:
6. 50 1 ; 8.00 3: : : : : : 4.50 4:
7.25 1 : 9.00 3: : : : : .: 5.00 12:
7.50 1 : 9.50 1: : : : : : 5.50 3:
8.00 4 :10.00 2: : : : : ,: 6.00 7:
9.00 1 : : : : : : : 6.25 :
0.00 2 : : : : : : : 6.80 2:
0 rec. 2 : : : : : : : 7.00 3:

: : : : : : : 7.30 1:

: : : : : ': : 7.50 2:

: : : : : .1 : 8.00 11:

: : _: : z : ‘ 9.00 1:

: : : : : ': = 9.50 1:

: : :_ : : : :10.00 ' ~:

: : : : = i 311.40 1:

: : = : = : =12.oo 1:

: : : : : = =1$00 4:

: z : : : : :20.00 3:

 

 

(5) EiTfi FROM REPORTERS.

One of the important pending developments in research in agri—
cultural economics is the expansion of our machinery for collecting
data of current changes and analysing them. The only agency that has
had extensive experience with collecting such data is the Division of
Crop and Livestock Estimates of the U; S. Department of Agriculture.

Its wealth of experience in this field would reQuire a year's work to
exploit and a book to present. All that can be done here is to indicate
the high points. -

The following excerpts from an article by Mr. W. F. Callander,
Chairman of the Crop Reporting Board, under.the title "The Crop Reporting
Service of the United States Department of Agriculture", outline the
principal features of the methods of obtaining the data:

(1) The township and county reporters, comprising one farmer re—

porter for each agricultural township in the United States, number
.slightly more than 40,000. Each t0wnship reporter is asked to report
monthly for the crops and livestock in his neighborhood, using printed
schedules, which, after being filled out, are mailed directly to Washingu
ton and are there tabulated and summarized. In arriving at state aver—
ages, the township reports are weighted by districts, each state being
divided into from 7 to 9.crop reporting districts. Each district is given
a weight in the state total proportionate to its relative importance.

'(2) Field Aids, of whom there are more than 40,000, are asked to
report the agricultural conditions within their observations and knows
ledge monthly, forwarding their schedules to the State Agricultural
,Statisticians in the various States. They are distributed similarly
to the township list.

Theoretically each township is represented each month by a re-
port from the township crop reporter, and presumably another report from
the field aid in the same township,-the first sending his report to Wash-
ingto and the second to the State Statistician. Actually many townships
do not have'a reporter on one list or the other, or perhaps on both, and
_ only about a 50 per cent return can be expected any one month from either
list. -, ' '

(3) Special price reporters, numbering about 15,000, report monthly
to the washington office the prices received by farmers for their pro-
duce. Another list of retail price reporters, numbering about 10,000,
are asked to report on prices being paid by farmers for seed, feed, machin-
ery, fertilizers, and other equipment and supplies. Both lists are made
up mostly of merchants and dealers.

4

(4) Mill and elevator reporters are asked to’report periodically""/
to the field offices on stocks of grain on hand, There.@IS-ahout-28+DOO
of these reporters. ,

(5) Individual farm renorters, numbering about 100,000llarslsent’
schedules on which they are asked to report only for their own farms.
Eiwir reports constitute.an actual farm census for.their individual

.1 5.13.3 .

0 (33
CD \.I

I
‘ K

, ‘Tho field offices have a similar list of 100,000 farmers which
th;y u“ .

intermittently.

_(7) Three times a year, once on acreage (September) and twice on

'snine and'other-live stock (June and December), about 200,000 reports
.91 individual farms are gathered through the rural carriers of the Post
Office Department.- The‘Semi—annual pig surveys_made as of June lst and
December 1st were inaugurated in the current‘hog production and intent—
ions as to future production by the individual sample method. Cards
‘ fare sent‘to‘each.postoffice in the United States, to be distributed by

‘ the rural carriers to the farms-along their routes, and direct to star
route contractors in sections where most of the rural‘mail is distributed
by them. . ‘ , - l

. -Each carrier is supplied with-ten cards, andis instructed to
‘distribute these among his patrons so that he will get returns from all
kinds of farms ~ in other words, a fair sample of the farms he serves.
He is‘instructed to get the information from the farmer himself and fill
out the Card, or else to leave the card with the request that it be
filled out and returned to him promptly. These cards when returned are
sent by the postmasters direct to the Department of Agriculture for tab-
'ulation.

In addition to the foregoing, a number of special lists, num-

‘ bering many thousands, are maintained of reporters on special crops such

‘ as cotton, tobacco, rice, potatoes, peanuts, truck and fruit crops, who
.report either directly to Washington or to the field offices for the par—

ticular crop in which they are interested and concerning which they are

regarded as the best source of local information.

The three major problems with which the Bureau of Crop Esti-
'mates has to deal are bias in the reporter's estimates or judgments,
selectivity in the sampling, and fluctuations in the sampling. Let us
consider them in the_same order. Mr. C. F. Sarle's* discussions of
bias in estimates followsz'

Where the farmer's chief source of income is

Bias in from one product, he is likely to be ultra conserv—
Estimates. ative in his estimates of yield per acre or condition.

This type of bias_occurs in a varying degree from
year-to year and it is not safe to assume that it is
constant. To some extent it is offset by the fact
that our reporters are usually better farmers than the
average. ' '
* Division of Crop and Livestock Estimates.

i":

I//
/

129' f

This ultra Conservative type of bias with yield;por-acre samples
may be much more important in a year of heavy production, when prices
are very low and there is a general dissatisfaction among farmers be~
cause of the low price. This kind of bias might be called price con-
scious bias. In the fall of 1926 the price of corn was low and there
was considerable newspaper publicity concerning large crops and low
prices. The corn yield reports from Iowa were fully two bushels below
the yield of corn reported to the assessor in the late winter and early
spring.

The last few years it has been our practice to send out a cotton
yield inquiry on March 1st asking for the yield of cotton for the last
crop. There seems to be much less conservatism bias in this inQuiry
than in those made earlier in the season, as invariably the yield is
reported higher than on December lst or earlier in the fall. Apparently,
as the crop leayes the farmer's hands, there is a tendency for the re—
ported yields to be higher. With grain crops in Iowa, the opposite
tendency is usually evident, due perhaps to shrinkage, or perhaps to
failure to remember, which may be called memogy bias.

Mr. John B. Shepard has summarized the experience of the Division
with memory bias as follows:

1: When there has been but little actual change from one date to
the next, there is ordinarily relatively little memory bias in reports
on most items, such as horses, cows, acreage of corn, wheat, etc. While
there are many individuals who do not report the correct figures for the
preceding date, the number who report too many is largely offset by the
number who report too few. Ordinarily the memory bias on wheat, hay,
horses, or cows is less than 1 per cent.

2: When a considerable period of time has elapsed, the errors of
those remembering too many is much less exactly offset by the errors
of those remembering too few, Thus, during 1927 the number of cows in
New York State increased slightly, and instead of the usual memory bias,
half of one percent too many, reporters remembered 1.8 percent too many.
During the same year the number of horses declined, and in reporting
for the previous year, crop correspondents remembered 2.1 per cent too
few.

3: Memory bias is much greater in items of minor importance, or
which change materially from month to month. For example, in reporting
poultry on hand a year previous, crop correspondents apparently remember
too few by 6 or 8 per cent. In the case of catch-crops, or others which
are grown only occasionally, the memory bias is often heavy,

4: Mbmory bias is much heavier in the case of items for which there
is no.exact definition. Fbr example, the memory bias on all corn as a
whole is probably between 1 and 2 percent in New York, but the memory
bias on corn harvested for green feed or fodder only may exceed 20 per
cent. The bias on oats is usually about one—half of 1 percent, but the
memory bias on grains cut green for hay is probably around 15 percent.

$5

130

5. On certain items there are certain twists of memory which are
SO frequently encountered that little reliance can be placed on the 7
hemory ef.‘ crOp reporters ”” flmong these are such items as dairy heifers
and other young stoC’k, including colts, lambS and pigs.

If there has been a freeze or a hail storm that has caused serious
m«ury to a crop, the reports from gTOVOla 1“‘11»ntly over—estiEate the
da:flg9 Ifweather cenditions are Ipazticularl; i1sronraging, the returns

a: e iikLly to reflect this temporarv d.i scourw~onr st. Fie .ld travel and
pois1nal contacts in producing aicas are esszntiaL 1f weather bias is to
bra a.Llowed for currently. ' - V V
The most significant aspect of this discussion is the tendency of .
the bias to vary from year to year, which might be discussed and illus—
treted at. greater length if Ispace permitted .1 1:
”.1 One of the most important uses of sample data,
Selectivity ~ ‘ 'is in connection with the determination of acreage.
of the Sampling*' A brief examination of the returns from almost any
inquirY'sent --to’the crop reporters of the Division
at once convinces us that the farmers who report to.
' " us are not representative of the entire group. The
individual farms reporting are larger in all states — Iowa, 288 acres
as compared with 157 for the census; Alabama 322 acres as compared with
76 for the census; Wiseens1n 155 acres as compared with 117 for the cen-
sus.

Along with differences in sige are other differences which may be
summarized as follows: ‘

l: The sample generally shows a hi. gher percentage of farm land in
crops.

2: The sample generally shows more head of livestock per farm.

3: Acreage samples generally show greater percentage of feed and
hay crops than census. In other words, ”there is distortion within the
sample. . ' -

4: Rural carrier surveys show less of this lack of representativeness
than mail returns.

5: Samples often fail of geographical representativeness — diSpro-
portion of returns on irrigated farms too .few larger plantations from
delta cotton region; too few“ from important tenant areas; too few from
marginal farming areas.

6: It can be argued a priori that condition reports are more repre—
sentative than almost any other data with which we deal. The factors
causing deterioration- in crops to a considerable extent prey alike upon.
the good and the poor farmer. The better farmers making- -up the crop
reporters of the Divi ion therefore represent all farmers insofar as
condition reports are concerned.

* Mr. J. A. Becker, Division of Crop and Livestock Estimates;

131

It must be remembered further that our condition reports are inter—
preted into yield per acre on the basis of past relations of condition
to yield. If there be a constant bias, it is cancelled out on both sides
of the equation.

7: The census gives a variable check on yields, but generally our
reported judgment yields exceed census yields somewhat.

8: Yields on the individual farms reporting'usually exceed those
based on judgment reports.

Mr. Sarle mentions some other interesting aSpects of this same
selectivity:

Any schedule that relates to a particular crop tends to bring out
replies from men who are particularly interested in this crop. It is
frequently desirable to have questions of general interest as a "lead"
on the schedule in order to secure a more representative sample.

The acreage surveys are highly selective because of the length and
difficulty involved in filling out one of the schedules, especially when
acreage for both this year and last year are requested. Only a high type
of farmer is capable of filling out some of the complicated inquiries
sent out by some divisions of this Bureml.

If there is wide—spread propaganda to reduce cotton acreage, the
replies to an acreage inquiry are likely to cone from the public—spirited
farmers who are actually reducing their acreage while the others failed
to report. This is frequently a very serious source of bias.

" It is fortunate," says Mr. Becker, "that our first
attempts to utilize sample data in the determination of
of acreage in crops and nwnbers of livestock were in connect—
ion with the change from year to year, Checking of the
percentage of change as shown on the two—year report a—
gainst such annual ezumorations as are available indicates
that while the individual farm reports are not usable as
absolute ratios, they are usable as showing the change from
year to year. In general, there is strong evidence that
the reporter’s change is rather representative of the change
of the entire group. Certain measurable biases have been
noted in the data, e. g., a general plus bias in the Northern
states, and minus in the Southern states, particularly with
reference to cotton. There is a possibility that the change
alert by certainxsizergrbupings may better represent the
change of all farms than does the entire Sample. A study
of his nature is now being made in the washington research
section"

132

The Question frequently arises in the minds of

Fluctuatiops ip research workers as to whether the general theory
Samoling.* of saiqling as developed by Karl Pearson, Yule and

Bowley, is really applicable to the sampling work

as done by the Division of CrOp Estimates. In any

given sample we have dispersion by various factors,
such as errors of observation, geographic differences, climatic differ-
ences, etc. The dispersion in the farm price sample of cotton in a
given state is caused by errors of observation, differences in price
due to differences in quality and grade, differences inprice caused by
distance to market, as well as differcrces in local competition. When
the whole United States is being sampled, the range in price for any one
of these factors may be very large.

The distinction between inquiries made periodically and surveys
made only once or twice, and then at no regular intervals, is also im—
portant. With periodic inquiries it is necessary to build up and main-
tain a list of correspondents or reporters to whom the inquiries are
sent. With periodic in tiries we are frequently fully as interested
in the comparative results of the two samples as in the results of the
inquiry itself. The very fact that we have the same list of correspondents
reporting for successive inquiries tends to strengthen validity of the
comparison between the s umles. With both the monthly farm prices from
price reporters and monthly crop reports from farmers, it is not at all
unusual to have 50 to 60 percent of the returns in a given month from
reporters who had replied for the month previous.

The type of frequency distribution obtained with a given sample
affects the dependability of the average to some extent when the dis~
tribution departs materially from a reasonably normal or symmetrical
distribution:. Fortunately a U-shaped distribution seldom occurs, but

_a J—shaped distribution is characteristic of the "grain stocks on farms"
"inquiry at the end of the crop year, and of acreage abandoned. Con—
dition or a drop in percent of normal is heldom reported much above
100 %; and as a result an average condition of 90 or higher is likely
to be obtained from a sample that is badly skewed with the “tail" to-
ward the lower values and only a few reports above 100. With such .
badly skewed distributions, it would probably be better if we could use
some other measure of the central tendence of the frequency distri—
bution than the arithmetic mean, but the time allowed for getting out
a report prohibits the use of any other. This tendency of the average
to depart from the mode is a serious matter with data that include
a large proportion of zeros, and should be taken into consideration in
connection with the use of such averages for making estimates.

The sampling method adopted by the Division of Crop and Livestock
Estimates is not that described in the textbooks as simple sampling, but
instead the stratification of the universe suggested by Yule and Bowley.
We stratify our universe into counties and crop reporting districts and
draw reports - observations — from each county. As a result, the fluct-
uations of sampling are less than if we trusted to pure random sampling
with our present facilities. In building up a list of reporters on crop
*The rest of this section is abridged from a report by Mr. Sarle.

155

yields, we endeavor to have one reporter in each township in the state,
at least several in each county. This is much better than if We took

the names of a few farmers out of a hat. It avoids the "bunching of
reports" that accidentally occurs with small samples and insures a better
geographic representativeness in our sarple, and in addition reduces

the probable error of the mean.

Not only do we stratify our universe on a county, and frequently
on a township basis, but we group several counties into a crop-reporting
district. The crep-reporting district average is then weighted by the
relatiVe importance of the particular crop in that district. In some
states, such as California, it is necessary to weight by counties be—
cause the conditions vary greatly betWeen counties. To the extent that
our crop—reporting districts are more homogeneous, that is, show less
dispersion, we are able to reduce somewhat the influence of the fluct-
uation of sampling by this weighting process. The practical application
of this Would lead to a special districting for each ihpcrtant inquiry
in-a giVen state. These districts might easily resemble the "gerry—
mander" districts of politics. We are already initiating a program of
re-districting the various states for the farr price work. This work
has already been done in Colorado and is underway in North Iakcta.

With the highly variable acreage sanples, it is desirable to group
our farms on the basis of the size of the.farm, using the census classi-
fication in order that farm area acreage Weights may be available for
weighting our ratios. The same is applicable with out inquiries concerning
the numbers of livestock on individual farms., In the state of Washing- '
ton it was pessible to re—district the state into three reasonably homo-
geneous areas. ' ' -

The acreage schedules are further grouped on the basis.ef the size.
of farm.} As a result of this size—grouping, or stratification,'the
disperSion of 4he size—group as compared with the district group Was
decreased almost enough to offset the fact that there were only about
oneefourth as nany farms in the Size group as in the district. 'The prob—
lem of dietricting a state for a special inquiry is of vital importance
in the future development of'this Sampling work. 'Wherever possible,
homogeneous areas should be made intb crop-reporting districts. It will
be necessary to increase greatly the size of the sample in the more
heterogeneous areas of the state. -

A further-nethod chstratification is to divide our question into
several subdivisions, the averages of which can later be combined'into
a figure for the whole by the use of special weights. Apple prices by
varieties and grades would illustrate this kind of stratification. Data
of quantity of sales-by varieties and grades would be needed as Weights

.with the prices by varieties and grades if an average price for all

apples sold was to be obtained.

A test of how Well either geographic stratification or question-
stratification really reduces the dispersion is to compute the standard
deviation of the data by districts to see if it is less than when'all
of the data are thrown together in the usual way or when present crop
reporting districts are used. Question districting has worked splenéfif
didly with the farm prices of beans and peanuts. .

‘134 - -;
\‘\\, .

Since a.great deal of our Sample data is used 1n,a:rolative sense
and the difference between two averages_is.really more important than
the averages themselves the probable error of the difference or of the
ratios becomes of prime importance. The‘actual computed figure for
probable error of a difference or ratio can be materially reduced by
the use of paired reports or identical reporters. The more identical
the sample and the larger the positive correlation between these re—
ports, the less the computed figure for probable error of the difference.
The actual significance of this, of course, is one thing when change is
being sampled and another when absolute conditions are being sampled.

Where averages are being used for comparative purposes, the thing
to do is to increase the number of paired reports from identical corres—
pondents for any two consecutive inquiries.

We can alWays decrease the probable error of our sample by increas—
ing the number of reports. This is very important if we are attempting
to make averages for small states or for areas smaller than a state.

‘ To what extent the sampling methods outlined
, Dispersion meet the conditions of simple sanpling and validate
the use of the measures of sampling based on the
assumptions of simple sampling, it is not possible
to say. It seems reasonable to assume that they
Justify one in believing that dispersion and the
size of the sample, the two components of Probable Error of the Mean,
affect the representativeness of the sample. There is a great difference
in the dispersion of sanple data depending on the kind of inquiry, whether
it be a sample of farm prices, yield per acre or acreage.- It is impor-
tant that the statistician know within what limits he can expect the
dispersion of his sample data to fall. The greater the dispersion, the
larger the number of observations or reports needed to produce a given
probable error of the mean. To illustrate, a sample of farm prices of
wheat in Kansas might easily have a coefficient of variation as low as
5 percent. With this dispersion, assuming the probable error formula
as entirely appropriate, a sample of 46 reports would result in a rela-
tive P. E. of the mean of 0.5 per cent. This is a highly stable aver-
age and not materially affected by the fluctiations of sampling.

If our sample had a diSpersion of 30 percent, to give us the same
relative P. E. of the mean of .5 percent, it would require about 1638
reportS. With a dispersion of 50 percent, about 4,550 reports would be
needed; with 100 percent dispersion, over 18,000 reports would be re—
quired, and with 300 per cnet dispersion, over 163,000 reports would
be needed. Unfortunately, however, there is a very definite limit to
the number of reports that it is practical to obtain by anything short
of a complete enumeration.

The farm price samples by states have the least dispersion of any
of the inQuiries made by the Division, ra.ging between 5 percent or less
and 10 percent vith farm products that are sold from surplus—producing
states and on well organized .orkcts, such as cotton, wheat, corn, flax,
and hogs in surplus~producing states. Apple prices are the most variable,
with a dispersion seldom less than 30 percent to as much as 40 percent.
\.

135

The dispersion of the retail price samples collected by the Department,x"
is seldom less than 8 percent,'even with staple products such as sugar,
and flour, and only with commodities such as socks and limestoneddoes

the coefficient of variation exceed 30 percent. .Farm wage samples show

a dispersion ranging from about 25 to 40 percent in most states.

The monthly condition figures may have a disPersion as low as 10
or 15 percent, and on the otherhnnd'as high as 30 or 40 percent in some
states. Even in the state of Washington, where there is a high degree'
of geographic variation, the coefficient of variation for winter wheat
in both 1926 and 1927 held at approximately 15 percent. Since condition
is expreSSed in percent of "normal" condition or normal probable yield
per acre, and since each locality can have a different "normal" concept,
it is to be expected that the condition samples would show much less
dispersion than yield per acre samples; and.that is the case.

Yield per acre samples seldom have a coefficient of variation less
than 25 percent, and frequently the dispersion reaches 90 or 100 percent,
v or even more, as-has happened in Oklahoma-where the cotton yields may
_vary from a few pounds to several hundred pounds per acre.

Land value figures show about the same range of dispersion as
yield per acre samples, as they may be as low as 25 percent in Iowa and
as high as 100 percent in California and some of the other western states.

Acreage samples show the greatest dispersion of all sample data un-
less it be numbers of livestock on farms. The dispersion of the acreage
per farm of the different crops varies from as low as 60 percent in Mary-
land to over 300 percent in the state of Washington. ‘

About 7,000 farm price schedules are new tab—
Size of . ulated each month from special price reporters, a
Sample list made up primarily of small town merchants, elev-
‘ator managers, and other locally well informed bus-
iness men. In the average sized state, the reports
tabulated vary from seldom less than 100 to 200, and
even more in the larger states. The number of reports is adequate for
the more important farm products and reasonably satisfactory even with
the price samples having the greatest disPersion. Where the product
is produced in surplus quantities, there are usually enough reports to
render the average reasonably stable and satisfactory. The areas of
surplus production have the heavy weighting in computing both the state
and United States averages of farm prices.

The regular monthly schedule to crop reporters is sent out pract-
ically in duplicate to the two different lists of correspondents. The
number of schedules tabulated from each list is seldom less than 100,
even in the small states such as South Carolina, and for the larger
stated, such as Iowa or Illinois, more than 1000 or 1200 schedules are
tabulated each month from the two lists taken together. It is only with the
really small states, such as Delaware, some of the New England States,
and the less populous Western states, such as New Mexico, Arizona, and
wyadtsg that the returns become so small as to seriously affect the
reliability of condition estimates, farm wages or the December lst farm

1.,

v

136

-prices of crops, and the January 1st prices of livestock. In the larger

'States, the returns are sufficient for reliable'state averages of yield
per acre.

V It is not until we takc.up the consideration of individual farm
acreage and livestock inquiries that the fluctuation of sampling be—
comes an extremely difficult problem. It is physically impossible in the
far western states where conditions are extremely varied to secure a
, sample of adequate size for acreage or live stock. In fact, UXCGPt in
the.Qorn Belt states and in some of the Southern states, the rural carrier
survey does not bring in enough returns to render the ratios dependable
as an indication of change. The fluctuations of sampling over~shedOW
the bias that may be inherent in the returns because of the selectivity
' of the sample.

The suggested remedy for this is the suggested annual census of
an identical sample of township. This sample census program is already
under,way in Alabama and it is hoped that it will be possible within a
few years to continue with this program in all the states. It will prob—
-ably be necessary to supplement the sample census with complete enumer-
ations of those particular crops which are grown intensively in limited
areas within a state. This will apply especially to the fruit and truck
crepe, Our present methods of handling fruit_and truck crops are in-
adequate from a sampling standpoint.

To summarize, the sampling experience of this Division indicates
that with most of the price inQuiries of the Department, the fluctuation
of sampli is not a serious matter. With yield per acre data, it is a
matter the must be given consideration. With individual farm inquiries
of acreage and numbers of livestock, the fluctuation of sampling is.a
definite limitation to our present methods, and we are suggOSting as a
proposed remedy the sample census of identical~farm areas. .A sample of
-from 5 to 10 percent of the townships in a state would probably be nec-

essary with such extremely variable data as those of acreage and numbers
of livestock. . '

While county estimates are highly desirable from the standpoint
of workers within a given state, they are not practicable from a statist—
ical standpoint without a very considerable increase of expenditure and
time.' The more variable the sample data, the less feasible are county
or district estimates. With a substantial increase in the.size of the
farm price sample, district piices within the state may be feasible for
the more important farm products. Grouping Several counties into reason-
ably homegeneous economic districts would help in this direction. There
should not be more than ten districts.in even our larger states because
of the difficulty of securing a sample of adequate size With yield per
acre, acreage and livestock samples. -It would even then be necessary
to limit the district estimates to the more important farm products.
Even assuming that the usual measures of fluct-r
Checks nation in sampling are adéQuate, no amount of analysis .
of the sample itself will enable one to measure the amount
of bias in that sample; It is only when check data are
‘ available that any measurement of bias is possible.. Some
indication of selectivity of sampling can be had from a study of thé'

137

of the distribution of the reports geographically, but it is very diff1~
cult to go much farther than this without having some reliable check
data such as carlot shipments or an enumeration of the particular uni-
verse to check against the average of the sample. Cotton ginnings, car—
lot shipments and the census and assessors' enumerations are examples

of check data.

The data supplied by cron reporters on schedules are supplemented
by other data based on personal travel and inSpection of growing crops
by the Agricultural Statistician in each state, on special field counts,
and interviews with well-informed men throughout the state, and by tele-
hone and telegraph when necessary. The Agricultural Statisticians also
accompany their regular reports to the Division with comments on each crop
in their states,'explaining conditions at date of reporting and stating
just what changes in the weather, soil conditions, plat diseases and in-
sect pests, and conditions of growing crops, have occurred since the last
report. .

Weather Bureau reports, as well as crop reports issued by various
state and private agencies, are used to supplement or confirm the in—
formation obtained by the Division of Crop and Livestock Estimates directly
through its own field service.

For use in making the January 1 inventory estimates, and checking
these and the pig survey and lamb crop indications, as well as giving a
basis for estimating livestock production by states, a large amount of
information is now being obtained as to livestock movements from and into
each state. This information is obtained from two sources. One of these
is from records of livestock markets, packing houses and similar sources
giving the number of head of each species received from each state and the
number of stocker and feeder animals shipped into each state. The other
is from records of railroads giving the number of cars of different species
of livestock forwarded from and received t each station in the state. from
Which county, district and state totals are obtained.

One of the most useful checks which have been developed in connection
with wheat is that of records of receipts at the mills and elevators, and
railroad shipments of grain. Because of the fact that with the exception
of the rather constant amount usedfor seed and feed, most of the wheat
crop moves into channels of trade, the Division is able to check its
estimates on this crop in a number of states.

watbhing how the checks work out each year, prevides tests of the
various measures of sampling. In practice, the effects of the bias of
the reports, the selectivity of the sampling, and the fluctuations of
sampling, never can be clearly segregated. Empirical tests will there—
fore continue to be relied upon in the final decisions. Egperiments with
the actual data of samples. such as that of Mr. Bradford B. Smith in
drawing 60 random samples out of 8000 pig survey returns, and of Mr. Sarle
with different crops in Iowa, will always be the safest guide to practice.
This paragraph is the most significant in this whole discussion. It
warrants amplification that space does not permit.

(6) SECONDARY DATA.

The examination of secondary data should include tracing it
through the same steps that one would go through collecting the
data oneself. This means that one should find out what units and
measures were used and how they were defined; and how the sampling
Was done if samples were used; also study_the schedule question by
question and the instructions issued in connection with it. Meth-
ods of tabulation and-summarization should be looked into if possi-
ble. Tables published by public officials often reveal glaring in-
consistencies if studied closely. The totals sometimes represent
different numbers of farms with nothing to indicate this. If data
from the same source are to be used over a period, special atten-
tion should be given to changes in units or methods of coiiecting
or handling data. '

It often proves possible to test the adequacy of such data by
checking it against data from.other sources. In some oases trial
studies are warranted. We need greatly to have a considerable
amount of census and other official data checked by re-surveys.
iny thus Will‘ve discover its weaknesses. When access can be had
to the original data of such sources, it can be tested by making
frequency distributions of a number of random samples from it.

The best discussions available of the uSing of secondary data
are in_Crum and Patton, 15-28; Secrist, 22-46; Jerome, 506-326;
Day, 589—393. Crum and Patton have a chapter on the "Compilation
of Secondary Statistics."

Whole chapters could be written on the nature, limitations and
methods of using the data of a single source, such as those of the
Bureau of Labor Statistics. An excellent study of this type is that
of Charles F. Sarle on the "Reliability and Adequacy of Farm Price
Data" (U.S.D.A.1480). C. R. Chambers discusses the accuracy of the
land value data of the federal census in U.S.D.A. Bulletin No. 1224,
"The Relation of Land Income to Land Values." We need similar
studies of the data of yields and production;of prices in the
wholesale markets.

The most important, and the most treacherous, source of im—
portant secondary data available to agricultural ec6nomists is the_
federal census. Most of the evidence points, as one would expect,
to under-statement rather than over-statement. The reader should
review carefully Dr. Joseph S. Davis' paper in the March 1925
"Proceedings of the American Statistical Association" and the dis-
cussion following it. Dr. Holbrook Working’s revision of wheat
acreage and production since 1866, in Wheat Studies of the Food
Research Institute, June 1926, should be read by all, first, as

.‘20

*1 ,'

'ovidence of the sort of errors that the census and crop estimate data con—
tain, and second, as an example of methodology in handing such material.
Some of the.principal types of errors in the census data are the following

1. Ommissions, duo in part to haste in taking the schedules.

2. Underestimates of total figures, such as sales, acroages, etc.
"iest personS'err in giving totals en the side of not appreciating
the size of the aggregate-~just as most people are unable to see

Whore their salary goes to each month unless they stop to itemize
~and add.

5. Conservatism.in reporting anything that relates to values and in-‘
come.

4. Failure to remember very exactly the events of the past year.
When the census was taken in April, the winter wheat crop re-
ported had been planted over a year and a half. There is usu-
ally a memory bias. For example, the cotton yields reported on
their own farms by 6000 farmer crop reporters to the Bureau of
Crop and Livestock Estimates in 1926 and 1927 averaged over 10
per cent higher in Larch than in the preceding December. f

5. Data from crop reporters indicate that 14 percent of the farmers‘
nwve each year, and other data from the 1925 census that twe~
thirds of these move betWCen December and March. This means
that the data given for those farms, especially if the census
is taken in the spring, if often given by a person who was not
on this farm during the year previous.

In using Wholesale price data, it is highly important to know exactly
how they are obtained, and the dangers inherent in the methods. Price re-
porting is still done on such an unstandardized basis that data reported
by different markets may rat be comparable. The system of quoting may
give a price which is not at all suited for the purpose in hand.

Most available secondary data purports to be a complete count or re-
cord; but when it does not, it must be examined with especial care from a
sampling standpoint.

X

(f) PREPARING DATA FOR ANALYSIS,

 

(1) Editing.

The principal purposes of editing work are as folloWs:

1. To reduce the schedules to uniformity in the matter of arrangement
so that they will be tabulated Without loss of time or error.

2. To complete the schedules by seeing that all totals and other on-
tries are made, and in some cases that the missing items are sup—
plied.

3. To check the data for accuracy and consistency of the various
parts.

4. To discard some of the data in many cases.

The proper procedure in the matter of editing depends in
considerable part upon the objectives of the study, whether that of making
a complete count or record, as in a census, or of presenting an accurate
description, or discovering_and measuring relationships.

Let us consider one at a time these four objectives of edit—
ing. The best procedure in the matter of reducing schedules to necessary
uniformity is for the editor to work over a censiderable number of sched—
ules to discover the changes which are needed, and to prepare a definite
set of instructions to be fellow0d. The following example of such a set
of instructions illustrates the nature of the problem.

(a) The apple trees reported as reioved during 1925 and 1926 should
not appear under "Record of apple trees" on opposite_page. In
case it is evident that they do, they Should be stricken out.

(b) In case trees listed under (b) (trees to be removed) are not on
opposite page they should be added.

(0) In case the variety of trees removed or to be removed is omitted
tabulate "Variety not reported" (Code 1%

(d) If report states no trees removed, list as such. But if-report
has no remark or note concerning the removal of trees it should
. be tabulated as not reporting.
(9) Under "times sprayed" edit to whole numbers for machine tabula—
tion; thus, if a grower reports part of orchard sprayed four

’ CF" a-

-v-- ..

-.._..

times and part six times, tabulate as five times.

(f) For machine tabulation, dusting will ordinarily be counted the .

same as spraying. With hand tabulation, dusting may be tabulated
separately if desired.

. There may need to be instructions about interpretation of
age reports, namos_of varieties, market classes, units of measurement.

Whether or not it is proper to fill in gaps in the data de-
pends largely upon the objectives of the study. If a complete count or
record is desired, there is justification at times for supplying a missing
figure on the basis of the average of the group or an average adjusted to
the special conditions prevailing. Occasionally only some procedure of
this kind will make poSsible the use of the rest of an important schedule.
The federal census formerly made use of this practice extensively. All
recent developments, however, have been in the direction of filling in
fewer and fewer gaps. Mr Leon Truesdell in his discussion of Dr. Davis'
paper on Federal Census (see 1928 Proceedings of the American Statistical
Association) I.stated that he believed that the practice should be still
further abandoned. The great difficulty with such a procedure is that the
cases in which an item is omitted are very likely to be abnormal, as when
COWS are reported with no dairy products sold, the cows cannot safely be
assumed to be producing an average amount of milk per cow.

If variations betWeen farms or other units or relationships,
are What is particularly to be studied, the filling in of gaps by applying
averages has the effect of obscuring the very thing wanted and reducing the
amount of variation. In a study of costs of producing milk based on cow
testing association figures, the costs of labor and several other elements
were included on the basis of average inputs. The authors were surprised,
to find, when they got through, that the Variations in cost between farms
were much less than ordinarily supposed, and drew from this the'suggestion
that probably much of the walk about wide ranges in costs is not warranted.

If an accurate picture of an area or region is wanted, as
in a survey study, there can be no purpose in filling in gaps becauSe do-
ing so adds absolutely nothing to the picture, since, at the best, the
completing is done on the basis of the averages for the rest.

Editing the schedules for accuracy is largely a matter of
checking titles and seeing whether the various parts of the schedule are
consistent with each other. It may be a matter of checking the opening
inventory plus purchases against the closing inventory plus sales. It may
involve matching valuations and acroagos, or division of real estate values
between land and‘buildings, or crop acres against sales receipts, or live
stock against food. In feeding studies, an important kind of checking for
accuracy is to compare the food consumed with feeding standards. Often-
times it is well to go outside the data themselves and obtain checks at
local creameries or elevators. Frequently the editing instructions should
include points covering checking for accuracy as well as uniformity and
missing items. .

The most obvious reason for discarding a schedule is its _

‘O_

/N

142

inaccuracy. The question is always pertinent as to how inaccurate
a schedule must be before it is discarded. If.only totals and
averages are to be used, it is frequently reasonable to assume that
a rather considerable degree of inaccuracy will prove to be compen-

' satory. It may be better in such cases to include all records, how—

OVer inaccurate, than to exercise judgment in excluding part of them,
because this very exercise of judgment opens the way for bias in the
editing. A better procedure is to establish some reasonable objective

'tests of degrees of accuracy, and follow those consistently.

The more difficult cases are those involving discarding
records for reasons other than accuracy. Oftentimcs some of them are
clearly outside of the universe selected for study and are properly
excluded. Even this should be done, however, only on the basis of a
clearly defined statement as to what the universe is supposed to in—
clude. Thus the field men making a survey of dairy farming may bring
in records from a number of farms in which the cattle are of the dual
purpose type and the sales of fat cattle figure very largely in the
livestock receipts. Should, or should not, these be included in the
analysis?

It frequently happens that a number of cases are af—
fected by abnormal circumstances. For example, a study of factors sf:
footing the price of grain making use of a large number of sales ac—
counts from records of commission firms may inelude a small number of
cars of grain which hare been noted on the sales record as "musty" or
"garlicky." Unless there are enough cars of those kinds to make pos-
Sible a special analysis of these factors, the proper procedure is to
strike out these cars. Suppose, however, a number of the records are
widely out of line with the rest and no notations of any kind are made
on them. There is good reason for believing that special circumstances
ware involved in these cases too, and no doubt it may be permissible
to discard them. Before this is done, however, a careful study should
be made of the circumstances which may have produced these exceptional
results.

A case of somewhat different nature is that of the single
farm, in a survey including perhaps a hundred farms on which the farmer
realizes large losses from having lightning strike his buildings just
at the time when they are not insured. t may be argued that this is
one of the hazards which should be included in an analysis of the farm
business and that therefore the record must not be excluded. The first
answer to this is that if a sufficiently large sample had been taken so

‘ as to give an unusual case of this kind its normal frequency, it should

be included, but not otherwise. A more pertinent answer is that inasmuch
as it is relationships more than averages and totals that is Wanted in
the study, the inclusion of these single cases furnishes no basis for
isolating the effect and the cause which it represents. Reasoning on a
similar basis, a half dozen farms on a lake shore might be excluded from
a study of land values in an area.

Then there is the question of discarding some of the cases
merely because they are extremely high or low. In an arithmetical mean,
these would figure very largely, and still more largely in a geometric .

.

\.

‘mean.~uln a mode or medium they have no more effect than the usual items.

If it was possible to say that the arithmetical or the geometric mean was
the scientifically accurate summary expression, than it could reasonably

be argued that extreme cases should not be excluded; but since such is not
the case, there is basis for the position that since they would not be in-
cluded if the mode or medium.Were used, they can be dropped, whatever form
of summary of expression is chosen. In correlation analysis, this question
is extremely important because of its use of standard deviation which accent—
uates the extreme. The difficulty of proceeding according‘to this reasoning
is that too much is left to the judgment of the editor. He is left in a
position where he may be accused of proving a case by selecting the neces-
sary data. The only protection against the exercise of bias, consciously

or unconsciously, in such a case is to lay doWn definite rules to be fol-
lowed.

It should be obvious from the foregoing that the person in
charge of a job of editing should have good judgment as Well as a sense of
scientific method. He must in addition be well informed as to the subject
of the study. ‘In one of the divisions of the Bureau of Agricultural Econ-
omics, the practice was followed for some time of excluding reports on
small ranches showing abnormally high incomes. Later it was realized that
many of these small ranches had access to a 12.rge area of public domain
which they grazed without any charge. In survey studies it is highly de—
sirable to have the man who had charge of the field work cooperate closely
in the editing; perhaps handling it alone, although there is something to
be said for having a colleague or superior give some attention to it also.
Oftentimes a wise procedure is to have some preliminary analysis made of
the data after they are edited, and en the basis of this do a certain
amount of reeediting. A simple correlation analysis may furnish the basis
for computing residuals for individual cases, which may reveal the fact
that some of them are seriously out of line and lead to a closer examina-
tion and checking of them.

Questionnaire data are the most difficult to edit. The ten—
denoy of careful workers is to exclude a very large proportion of such re-
turns because of their apparent inaccuracy and the wildness of their esti-
mates. This very process, however, may at times introduce a more Serious
bias of one sort than it eliminates of another sort.

Any large job of editing should be carefully organized, and
the more obvious and glaring of the mistakes, omissions and inconsisten—
cies may be detected by clerical assistants. The steps can be as follows:

1. Have the clerks bracket all doubtful answers.

2. Have the head clerk red-pencil these which in his opinion are valid.

5. Have the editor check the work of the head clerk, perhaps using a
blue pencil.

With the Work organized in this way, one editor can pass
final judgment on many thousands of schedules in a short time.

144

(2) Checking For Accuracy.

. . The following are some rules or procedures to be followed
in checking for accuracy:

1. Have the clerks always initial their work.

2. Have the second clerk check the work of the first, both initialing
the sheet;

5. Have all tables arranged so that they can be cross totaled; also
have tWO persons independently calculate all totals and averages.

4. Keep all related shoots properly fastened together.
5. Do all checking for errors before making the summary of the totals.

6. Use different colors of ink for various classes of items; for ex—
‘ ample, red ink for totals, green ink for grand totals.

7. Watch cyphers and decimals very carefully.

8. Have all transfers or entries checked by some other person than the
one who made them in the first instance.

(3) Tabulatign.

The whole subject of economy and efficiency in tabulation
work is one which will well repay careful study. All that can be done here
is to make a few suggestions which have greWn out of the experience of work-
ers in the Bureau of Agricultural Economics and a few other places.

In many types of studies, time Would be saved if more of the
tabulating work was done either from the original sheets or from office
sheets compiled from them. The plan of reducing field data to a compact,
well arranged office sheet on a good quality of paper, such as described by
Prof. hisner for handling farm management surveys, has stood the test of
experience and has a great deal to commend it.

If no correlations of any kind are involved, it is frequently
good economy to work up each item independently directly from the office
sheets. If correlation between two series is wanted, then these two can be
Worked up directly from.the office sheets. If a considerable number of re-
lationships are involved, it is advisable to get the data for these on pro-
perly arranged cards so that they.can be sub—sorted. It is easy to encum—
ber such cards with too much data and try to analyze too many relationships
at one time. If desired, regression results and residuals can be entered
on these same cards.

The plan of transferring all the data to long strips has not
grown in favor. The strips have proved to be too cumbersome and difficult
to Work with, especially by those not entirely familiar with them.

There is an increasing tendency out in the states at the

\—

:3.“

[.l
4;.
I

\.

present time to use sorting and tabulating machines. It will be interest-
ing to research agencies generally to know that the workers in the Bureau

of Agricultural Economics are using such mechanical aids much less than
they‘wero, and that the Division of Crop and Livestock Estimates has ceased /
their use almost altogether. The consensus of opinion in the Bureau of
Agricultural Economics on this point is as follows: It is advisable to use
mechanical tabulation when the number of cases to be tabulated is large and
when a considerable number of sub—sortings or cross tabulations is wanted.
It is especially economical to use a machine for tabulating when data of the
same type are brought in at frequent intervals so that a definite system

of coding can be standardizid and learned by the clerks.

An important disadvantage of machine tabulation is the large
number of errors made in the punching of the cards, especially when this
punching is done by clerks who are not familiar with the project. Many Who
have used machines have come to the conclusion that they can be safely used
only where the schedules have been very carefully edited, so that they are ,;
fool-proof, and that even-then all punching must be supervised very care-
fully. There is the further difficulty that errors made in the punching ~
are very difficult to locate.

In studies Which lend themselves to machine methods because
of their siZe or the large number of sortings, the COLL can frequently be
reduced as much as one-third. The saving in time is really in the sorting.
Nr. Becker says that only in studies involving considerable crossatabula-
tion does machine work pay. One unit consisting of one tabulator, one sort-
er, and punchers and vorifiers costs, at present retQS, $175 to $200 per
nmnth (the manufacturers rent their machines) and there must be enough Work
to keep the rechines fairly busy if the total expanse is to be less than if
the work were all done by hand.

146

- 1g) ANALYSIS OF DATA.

(1) Summarizing.*

It is the purpose of the handbook in handling such statis-
tical devices as averages, frequency distribution, and measures of disper-
sion merely to show how they may be significantly used in agricultural
economic analysis. To the young research worker, these devices too often .
seem largely academic, reminding of laboratory exercises in histograms,
polygons and ogives of statures, leaf measurements, etc.

For frequency analysis, three sorts of uses will be consid-
ered: (1) Description of a group; (2) Comparison of groups at different
periods or in different places; (3) ShOh'ing relation between variables.

Group variations Two of the striking uses of frequency analy- "
sis for purpose of description of a group are the
follmVing.

l. Skew or "bulk-line" phenomena in variations in some items of cost
or physical input or profits or income.

2. Behavior of receipts, sales and prices in the market place.

Suppose we wish to examine the quotations of the prices of
a product in a particular central market as a representation of the actual
sales of the product in the market on that day. The quotations of the
Bureau of Agricultural Economics for July 20, 1922 were as follows:

Canner COWS $2.00 — $2.50
Cutter COWS 2.50 - 3.25
Common to choire cows 3.25 - 7.50

Table I shows that there were no actual sales between $5.50
and $7.50, which indicates that the Quotation on "common to choice" cows
must have been nominal.

 

* Professor Warren C. Waite, University of Minnesota.

TABLE Io

“Distribution of the Price: of Slaughter COWS Bought by Swift and Com-
pany and Armour and Company at South St. Paul, on July 20, 1922.*

 

 

 

Price per Per cent a?
Hundredwaight Animals purchased
$1.88 - 2.12 0.8
2.13 - 2.37 7.3
2.38 - 2.62 19.8
2.63 - 2.87 11.2
2.88 - 3.12 15.5
3.13 - 3.37 11.2
3.38 - 3.62 9.1
3.63 — 3.87 8.1
3.88 - 4.12 4.9
4.13 - 4.37 0.5
4.38 - 4.62 5.2.
4.63 - 4.87 1.0
4.88 - 5.12 ~V4.2
5.13 - 5.37 0.2
5.38 — 5.62 1.0

 

* Gaumnitz E. W.,

Central Market Price Quoting, Ph. D. Thesis at Minnesota,

 

1925, p. 104.

. ..._.. - -1.--- ”

TABLE II.

 

‘

A Random Sample of 5,934 Hogs sold at South St. Paul on June 1, 1926.*

 

 

Price per Per cent of
HundrodWeight Animals purchased
$7.38 — 7.62 3.88

7.63 — 7.87 3.62
7.88 - 8.12 4.16
8.13 — 8.37 11.85
8.38 - 8.62 9.43
8.63 — 8.87 11.19
8.88 - 9.12 48.70
9.15 - 9.37 7.15
9.38 - 9.62 0.60
9.63 — 9.87 0.50
9.88 -10.12 3.10

 

*‘UnpuBlished data in files of’Division of Agricultural Economics at the

University of Minnesota.

Table II gives a similar distribution for the prices of
hogs on the same market on June 1, 1927. The difference in the nature of
the two distributions is striking. The quotation "average cost to packers"
was $8.60 on this day. '

One wishing to know what actually happened on the marketéfin

148

a given day in the matter of prices can learn a good deal by making a fre-
quency distribution of the sales, and still more if he has some sort of a
normal for that time of year and day of the Week to serve as a criterion.
If the buyers have acted abnormally, it is pretty certain to show in the
distribution of prices.

The art of quoting prices is largely based on the science of
frequency distribution. The job 0 the price quoter is to find a figure which
represents what a normal array of the commodity, or of a -given class or
grade of the commodity, would have sold for on the market that day. He
has to determine this in many cases out of an imperfect distribution in the
first place, and almost always out of an insignificant samplo'even of what
sales are made. Those.who use market quotations need to know-the proce—
dure that was followed in making them. Some day this procedure will be re-
duced to a more nearly.soientific basis. Scientific grading-awaits such
developments. ' ~ -

The shape of an individual frequency distribution in itself
often meansI little. But as time goes on, we shall build up an understand-
ing of the "normal" for various phenomena. Stability.cf- frequency dis-
»tribution of samples it a safe guide in any case.

Such use of frequency analysis is especially liable to the
errors inherent in choice of clasinntervals. IThe essential principles
determining choice of class interval are covered in the textbooks.*

An interesting example of a cumulative frequency distribu—'
tion is given in Figure 5 of the U. S. .Tariff Commission‘s final summary
oft the cost of producing sugar beets.

The frequency analysis Imay also be used to indicate cases
in which a further classification of our data may be useful in an analysis.
According to the Le Place - Charlier hypothesis, any frequency curve may
be thought of as a sum of independent frequency curvesnl‘”I A study of Sales
of hogs in the St. Paul. market on June 50,1927 gave an example of a dis—
tinctly bi-modal distribution. hhen broken up into its. proper.parts, this
distribution gave a classification by weights corresponding to. that used by
the Bureau of Agricultural Economics in quoting hog prices-.

Comparis cons There are many economic problems.which are
0 Frequences. . primarily concerned with differences in the dis—
_.____....____

. tribution of a particular variable.r These dif—
ferences are sometimes related to- geographical
differences in sizes of farms in different sec-

tions of a state, or the amount of net rent of farms absorbed by taxes in
various states, or wages of farm labor, or farmerIs incomes, or intensity

 

* See especially-—Yule, G. U., Theory of StatiStics, p. 79—83.
Day, E. E., Statistical Analysis p. 65—71.
Mills, P‘. 0., Statistical lethods, p. 68-73.
** Fisher, A., Mathematical‘ Theory of Probability, p. 176.

l .r -~

w

of cultur€, Another kind of study deals chiefly With diffs
standpoint of class or kind.

‘ different nationalities on farm living.

I" O

“I

no... ..

renees-fronrthe

Thus we may wish to compare the variation
of incomes of tenants with those of owner-farmers, or the expenditures of

Finally, there will be-many cases

where the significant thing to be analyzed is the difference in the varia—

tion at different periods of time.

We may, for example, wish to analyze

the changes in the distribution of income over a period of time, or sizes
of farms, or certain aSpects of market behavior, progress of settlers,

changes in marketing costs, and so on.

Frequency distributions offer a method of attacking such problems.
A separate frequency can be prepared for each geographical area, claSs
or type, or period of time and these compared asva series of frequency
distributions. The comparison is facilitated if the class intervals
are the same in all frequencies, and whereever possible these intervals
should be chosen at the start of the problem with such comparison in
view. Whenever the frequency distributions differ materially in size
and shape, and there are a great number to be compared, it will be ad-
visable to compute certain coefficients descriptive of them and compare
these, such as these of control tendency, dispersion, skewness or
kurtosis. Another useful method is to express the data in percentage
of cases falling in each class, as in Table III.

TABLE III.

~ Percentage of Farm Operators Falling in Certain Age
Groups by Geographic Divisions. '

 

‘hs.

: Under

:25-34 :55-44 :45—54

:55-64

 

:25 yrs“

1.69

New England :
Middle Atlantic : ‘2.52
East North Central : 3.55
West North Central : ~4.87
south Atlantic ; ; 7.26
East South Central : 8.93
West South Central : 8.98
Mountain - 3.95
Pacific

2.67

rycars :years :ycars

: 11036:
: 14.89:
: 19.73:
: 24.29:
2 19.90:
: 21.54:

: 23.47
: 24.16

17.01:

20.72 E

23.05

25.81

: 25.61
24.90 :
26.23 :
24.69 :
24.09-:
x 25.40 :
: 28.00 :
26.16 :

24.32
22.59
23.26
22.51
22.22
22.83
25.52

:65 years
igyoars :& over
: 22.19 : 18.22‘ ,
E 20.42 : 13.51 ‘/*
: 18.01 : ~9.49 4 ' "
: 14.87 : 7.14' , .1].
:‘15-09 : 9.80 ' ‘ ,— ““
: 13.77 : 9.16 ‘ w ”
: 12.81 : 7.12
: 14.61 : 6.45
' 19.05 : 9.59

 

\_ ‘
.1Teble IV 13 “ °°mparifiop~Ih9re~di£1erence in frequency and

overlapping of frequencies isximPOrtant. Theleifect of dressing per-

' centage upon the grade of the animal is of course-evident.‘ The

extent of overlapping indicates the existence of other factors than
dressing percentage in determining the grade of the animal. or that
there was a shift in the kind of animals classified in the particular
grades during the period. There_is also a difference in the shape

of the frequency distribution of each grade, which is importent for
certain aspects of the problem.-

TABEE IV . -._ '- X

-

Cows Bought by Swift and company and Armour and Company at Sonth St. Paul
during the Period July 13 to September 15, 1922, Classified According
to Grades and Dressing Percentages.‘

 

Dressing Percentages

 

48: 49} 50: 51: 52: 53554z55zNoAn~

 

 

Grades S 39: 4O : 41 : 42 : 43 : 44 : 45 : 46 : 47 :
: to: to : to ; to : to :_to_: to : to : to.: to: to: to: to: to:.to:to:to:imal
: 40: 41 : 42 : 42 : 44 : 45 : 46 : 47 :‘48 : 49: 50: 51: 52: 53: 54:55:up:g
Banner §'97§1746§3321§2907§ 755; 340: 76: 97: 7; ;- g 45; Q Q 2 Q9409
Gutter :108; :8; 189;.793;1656£2258:2483;1674; 520; 74; 42; 41; ; 43; 8; I; g9867
30mmon ; g 16; g g 37; 227; 390;1070;1818;991;579;135;126g ; 3; ; ;4892
Jedium § E Q E Q 6; i 4; 44; 24i124§255§213§143§ 55; 25; Q12; 915
Eood Q E § § Q i i .i 1: i i i g: 17; : § § 3; 20
Jhoice ; Q g i i i i i i i : i -: : ~ ; g «t- i
' Data obtained from records of Swift and Company and Armour Company.

(I

of

western University has developed some interestingflgnnnlupions out of'suoh
an,ana1ysis of retail stores.

Comparison of the.
in a frequency distribution from your to throws light on the theory
"representative _fiarms” of Alfred mar

Correlation betWeen
Variables

”151

n of individual farnsgnasmwvnwntg

Professor Socrist of North-

\

Problems which involve an httgrqd to relate the
variation in a particular variable to the varia-
tion in other factors are usually handled by a ,
correlation analysis. A frequency analysis, how-
ever, is often advisable as a preliminary step to

this or may serve as a substitute in problems which are relatively simple
or do not contain enough observations to warrant the use of correlation
methods.

The method of approach to such problems is by the prepara-
tion of a double frequency or contingency table such as in Table V. This
table indicates that in- general the smaller farms have more cows relative
to their crop area than the larger farms.

TABLE V.

Dairy farms classified according to the number of crop acres
- and the number of cows kept. *

 

 

 

 

Number of: Number of F arms :All
cows :16 t0.131- to: 46 to: 61 to: 76 to: 91 to: 106 to: 121 to: 156to: 151 to: 166to: sizes
30 z 45 :v 60 : 75 90 : 105 : 120 : 155 : 150 : 165 : 180 :
:acresfaeroszacroszacres:acres:acrosfacros :acros :acres:acros :acros: :3

5 orlnss: l : 2' --—- :---- :-----:---- :—--- :—--— :---- :---- :---L : 5

6 to 10 : 12 : 32 20 . 7 : 1 :--—- : 1 :---— :---— :——-- :—--- : 73
11 to 15‘: 3 : 19 25 : 21 10 : 4 : --------- ———- —--— --——— : 82
16 to 20 :-—-- 6 23 : 16 . 16 s 3 z 1 ;-_-_ ;____ ;--__ : 71
21 to 25 :---— 2 5 ° 3 : 7 6 : 1 : 1 : 1 :-——- :-—-- z 26
26 to 30 :—-—— -—-- 2 1 3 4 , 2 :—--- ' l :—-—- :---— :---- : 10
31 to 35 :—-—- ----- ---- :---- :—--- :---- : l :-—-- :---- : 2 :—-—— : 3
36 to 40 :—--- ---- '—--~ :-—-- ------------------- __-- --_- 1 : 1
41 to 45 :---— ---— ---— ._--- :---— :~—-- ;---- ;_--_ ;-__- :____ :____ :___
46 to 50 :—--- ---- 1 --—- ---- ---- :—-—- :---- :--—- :---- :-—-— : 1

Total : 16 61 76 : 48 : 38 : 18 : 6 : 3 : l : 2 : l :270

 

L

The analysis may sometimes be concerned not alone with the
association of the tWO attributes of the variable, but with some result of this

* U. S. D. A. Bulletin 1400 — Factors Affecting Farmers Earnings in South—
eastern Pennsylvania, p. 10.

~~‘.\

152

association, as in Table VI.n Here we are interest-ed primarily in the pro—
fits of stores, which are a result. .of the relation between total oxpthos
and gross margin.- The class whore Ithoso .are approximately the same is“
boxed by heavy. lines.. Sections to the right and above indicate 1ossos, '”
sections below and to the left indicate profits.

TABLE VI.

Total Esponses and Gross Margins of Ono-ham meat Stores in
. Chicago, Cleveland and Nevr‘York, wk

 

Gross Margin : I Total Expense — Per cent of Sales

 

 

 

Per cent : 14‘; 16 : 18‘: 20': 22 5'24 : 26 : 28 : 30 : 32 3
of Sales : tn?: to ; to : to t to = to : to : to : to‘: and : Total
. : 16 :-16 : 20 : 22 2 24 c 26 :‘28 z 30 1 52 : over 6
Under 14 :-—--: 2 : 1 : 4 : 2 g 3 : .2.: .1 z 2 ;:-,3- :w 20
14 to 16 z 1 :-——-: 5 : 2 : : z : l t 1 1 : 8
16 to 18 : :—-—-:—-1—: 1 : :. t 1 : : 1 2 4
18 to 20 : 1 : 1 :----:——4-: : z : 2 : 1 : 1 10
20 to 22 : " l . . 2 : 1 : l : 3 : 2 : V -: l 11':
22 to 24 1 : = .. :~-2—_:--3-: 3 : 2 : 1,: 2 14
24 to 26 2 : l 2 7 : 2 :--2~: -~4—: 1 : : l : 20
26 to 28 : : 1 : 1 ' 5 : 5 : 2 :—-1-:-—5-: l 21:."~ 19 m
28 to 30 ': '1"' = : 2" ‘4': 2m: 2 E—-2-: ——3—: 111 ”E 16
30 to 32 : : : 2 : z 4 : l ' 3 z 2 :--1-:---1~—: 14
32 a over : : : 4 : 6 z .4 : 4-:.15': : 1 :-g-5—-: 27‘
Total 2 : 2.: s g 13.: 55': 24~z 18 a 22 :;1s : 12 : 15: : 163~

 

* Bureau of Business Research, Northwestern bn1v0r31ty, :Serics III, No. 9,
"Expenses, Profits and Losses in Retail Meat Stores," p. 65.

_ The analysis may be carried further without the use of cord
relation by the calculation of certain coefficients of. association and con-
tingency. The analysis with respect to these methods has received treat—-
ment in the section on "Association."

Averages 1 . Since an average is a-summary or characterization
.3. _ of data by a_singlo figure, it.is not useful Where
' the,significant thing is the-variation of the data
rather than .a s.ingle type characterization. Thus
an average cost of the production of wheat is quite
meaninrless if we have in mind the use of that average to guarantee the farm-
er a certain profit above cost, or propose to set a tariff upon the basis
of it. The significant thing here is the difference in costs of the various
producers which precludes the establishment by tariffs or otherwise of a ”
single price that will guarantee a profit to all. There are many similar
situations. - -

Averages fall into two groups; thoSO Wh1Ch depend for their ”
value on the magnitude of.each item of the series of calculated averagesl.

$.
‘6

\ _--

‘- “v/

such as the arithmetic mean; and vahler‘fliich_,dopnndm‘tho~mmber or
position.ef certain items, such as the mode or median. Tho former,
since they depend upon the value of all the items of the series are
subject to algebraical manipulatien.' They can be combined with similar
averages from other groups to give the same average for the combined
group as Would be obtained from a similar average of the original items.
Albegraic manipulation is an important attribute of an average; in
fact, Yule considers it the most important single attribute.*

The arithmetic mean is the most widely used of all the
averages. It is rigidly defined, based upon all the observations, rea-
sonably easy of calculation, lends itself readily to algebraical treat-
ment, and is intelligible to the layran. All things considered, it is
probably the most useful average, but as is pointed out later there are
many circumstances in which its use is not justified. Examples of the
arithmetic mean are no numerous that no specific cases need be given.
It is used almost exclusively in the census, to summarize surveys, and
in the yearbook.*# ‘ -

The geometric mean is the n22 root of the product of the
numbers comprising a series, the n signifying the number of items com—
prising the series. It is also 1353 natural number corresponding to the
arithmetic mean of the logarithms of the items comprising the series.
It is thus a form of an arithmetic mean. Similarly to the arithmetic
mean, it can be determined if we knew the total product of the items
and their number without knowing any of the individual items which com-
prise the series. Just as the sun of the deviations from the arithme-
tic mean is zero, so the sum of the logarithmic deviations from the log-
arithm of the geometric mean is zero. This means, in terms of natural
numbers, that the sum of the ratios to the geometric mean of the values
under it is equal to the sun.ef the ratios to it of the values above
the geometric mean. In other words, we have a mean in which an item
fifty per cent of the size of the mean receives equal importance in the
determination of that mean with an iten.twice the size of the mean.

These properties of the geometric mean make it an impor—
tant average for rany sorts of summarizatien. One type of these prob-
lems is where We are averaging rates of change and wish to give equal
Weight to equal ratios of change. The simplest example of such a prob-
lem is the determination of the average rate of growth of a population.
Suppose that the agricultural population of a country has doubled during
a cried of ten years, probably at irregular rates. The geometric mean

’0 2 1.0718, gives us the rate at which the population would have had to
increase regularly to produce the result which we find in the tenth year.
Thus 7.18 per cent is the average annual rate of increase. We are able

 

* Yule, G. U., Introduction to the Theory of Statistics, p. 108

fi'A comprehensive discussion of the adequacy of the arithmetic mean,
particularly Weighted for farm-price summarization, will be found in
U. S. D. A. Bulletin 1480, Reliability and Adequacy of Farm—Price
Data.

154

from this result to estimate the population for any periodni‘ Many
problems of this sort are complicated. For example, Sewall Wright in
his study of the relations of corn and hogs found it necessary to cor-
rect the estimates of corn acreage of the Bureau of Crop Estimates
from 1870 to 1889, because of the breaks in these estimates when ad-
justments were made to the census figures. The acreages in the inter-
vening years ware estimated by swinging the fluctuations shown by the
Bure.u of Crop istimates botWeen the census figures, by multiplying
the rte io of estimates for successive years by a constant factor to
make the tenth year agree with the next census. **

The second general use of the geometric mean is in aver—
aging ratios or relatives when these ratios or relatives express vari-
ations, that is, increases and decreases. This is the problem with
which we deal in index numbers. Here the property of the geometric
mean of giving equal weight to observations, say of: seventy-five per
cent of the mean and one hundred thirty-three -per cent of it, becomes
of great significance. This use of geometri‘c means will be developed
in the section on index numbers.

The mode represents the most probable single value of
the variable, and there is usually a clustering of a considerable pro-
portion of the other values around it.*** It represents in consequence
a very important characterization of data. Its value is unaffected by

 

* The population at the end of any particular year in the period may be
may be determined by the following formula:

P (1 + r)“;

Where P I: population at the beginning of the period, r: the geometric
mean of the rates of increase, n: the year under consideration and P1
3; population at the end of the year for Which the estimate is made.

** The formula for any given year was as follows:

El E.C [o
E :1 2»
10E

where xsthe year for which the value was to be calculated Cl, C2: The
two census figures between which the crop estimates were to be swung.
E1, E2 : The original estimates for the same years.

*** In some cases, two or more points of concentration may appear.
These may be due either to the small number of cases, to inclusion
of data properly belonging to separatn frequencies in a single fre-
quency, or in rare cases to underlying causes even where the data
properly fall in the same frequency analysis.

unusual observations. The mode is therefore the proper average to use
when We are interested in the typical size of the individual observa-
tion. In a study of price quotations we might, for example, compare

. the bulk-ef-sales quotation with the modal sale. Ifl this sale lay out-
side the bulk-of-sales quotation, as is sometimes the case, we would
conclude that the latter was not a satisfactory representation of the
true sales. We would also be interested in the typical or medal in
comparing the incomes of classes of farmers and city folk. In the same
way, comparison of the size of farms in different states when We are .
thinking of typical farmers should be made by modes. The mode is the
average to use when We have in mind individual cases, the arithmetic
average When totals or aggregates are the significant thing.

The harmonic mean of a series of quantities is the re?”
ciprooal of the arithmetic mean of their reciprocals.* The harmonic
mean finds use only in a restricted field, but in these uses it has a
distinct advantage., it is used in averaging prices in cases where
-they are expressed‘as "se‘many per dollar" and the average number‘puré
chased per dollar is reguired. The average number of a group of come-
modities purchased per dollar‘ebvieusly depends upon the quantity.ef

each purchased at-these different rates.~ Hence harmonic means may-need
to be weighted.** ‘

The median is the middlemost or central value of.a series

when the values are'ranged in orderlof magnitude. One of its'principal
uses is in-cases where the items effa series.are not capable of quanti-
tative measurement, but can nevertheless be ranked. Thus in a study of
the attitudes of farmers tOWard a particular thought or object, We might
be able to rank'those attitudes without definite measurement and discern
a median attitude; The second principal uSo_of the median is in cases'
where We Wish-definitely to avoid influences of unusually large or small
items. Percens has recommended the use of the median in his method of
link relatives for determining seasonal variation.*** The Harvard
Business SchOOl group has also used the median extensively in studies of
the costs of doing business in various types of establishments and'lines
of business.****-' I -

. ' . When the distribution becomes asymmetrical, the arithme-
tic mean, the median and the mode do not coincide. The mode lies far-:
thest in the directiOn of asymmetry and the median betWOcn it and the~
arithmetic mean.- If the distribution is fairly'normal, the arithmetic
average appears'te be the proper average for summary._ It has approxi-
mately'the same value as the median and mode and approximates their _
characterization of the series, and has the advantage over them that it

., ; _. .1 .

* _l_l_..._-1 2(1)_ {(i) "

**.An example of the use'of the harmonic.moan is found in Minnesota Tech—

-. nieal Bulletin No. 10, "Factors Determining the Price of Potatoes." .

*** Cf. Reitz, Handboek‘of Mathematical Statistics, p. 151—158. ,

*ffr A_similar usegin agriculture may be found in U. S. D.‘A ;*Depfi£f'”"
Technical Bulletin 13, Cotton-Gin Operation in North-Central‘Texas,
'po'13. ' ’ " ‘ I . »

 

 

1 .
v

156

\’
can be built upon. If the distribution is not ndtgal, as is usullly
the case with the economic and social data, then S no other average may
prove more useful. If the analysis is to step with the particular dis-
tribution and not be built upon this, the choice will usually be the
mode. When the distribution is skewed toward the small values, and it
is desired to proceed beyond the distribution in the analysis, the geo-
‘metric mean will probably be chosen. This ehoifip of course Will not be
made when the mode lies to the right of the arithmetic mean. Greater
care is necessary in the choice of an average to characterize an asym-
metrical distribution than one which is symmetrical. -

'Dispersien It is advantageous in many cases to have a more
- complete description of the data of an analysis
than is previdod'by averages of series. This
description may be provided by certain meas—
ures of variability. These may be used either
to give a more complete description of a partieular'serios than is prob
vided by its averages, or to provide a means of comparing the variation
in tw0 or more distinct series. The latter problem, of course, demands
the more generalized measures. .

_ Certain types of problems depend principally upon the
question.9f whether one series sdeS a greater degree of variation than
anotheruquor example, We may Wish to know whether there is a greater
variation in the size of farms er in the income from those farms. Again,
we may be interested in the relative amount of fluctuation in two series
as determiners of a result which we find in a third series. For exam-
ple, we may wish to know whether acreage or yields fluctuate more in
order to :udgo their comparative effect upon crop production which is a
product of the two. Variations are also significant in a wide range of
problems involving sampling.

The problem involved in the determination of the varia-
bility of a time series is quite different from that involved in the
comparison of the variability of a series of geographical data or phy-
sical measurements. In the latter, we are ordinarily concerned with a
comparison of the absolute values of the variable. In tho_former, our
interest usually lies in the variation of one observation from the one
immediately preceding it, or bethen groups of related observations.
In a time series analysis we do fininthh variation of data adjusted for
changes in the price level. This adjustment may take the form of de—
partures from some trend line for the data. ' '

There are two principal ways in common use of measuring
the dispersion of a series.* The first type consists of some measure
of the difference of certain counted items. Here we have the range,
the interquartile range, interperoentile ranges, and so on. The second
type consists of some measure of the difference between a certain aver-
age and each of the actual items comprising the series. The average

 

* Fitted frequency curves Would give a complete description. ‘Tho math-
ematical technique necessary for their fitting, however, prevents any
wide use.

M. NH.

,._,~ «‘ _,.._.. _ .-

.. .— >- ...\

, —-'- ‘ r - w \ \. lax
‘ _ _ /.

’*\\ -dov1atien and the'standard deviation  are,examp1o5,c£/:his type. §ini—~

lar m0th0d5~0r0 f0110WQd11n -doterndning_measnros for the comparison be-

tween series, except that~the_lmb§ar must.bngednood to a comparable

basis.

.>~H.‘,_. .—

*“\—..

“Kai - ..{

\

The simplest measure“ the first type is the range of
the series. Yule says of the range:*» ~hqsi‘mplest possible measure of
the dispersion of a series of values of a variable is tho-actual range,
i. 0-, the difference between the greatest and least values observed.
While this is frequently quoted, it is as a rule the worst of all pessi—
blo measures for any serious purpose." It gives us no idea of the form
of the distribution between these extreme values and is very erratic,
depending as it does on infrequent values at the ends of series. The
range gives an idea of how wide the variation may be, but no indication
of what may be considered a typical variation. It can be used to ad-
;yantage principally for such things as the range in costs of marketing

' operations ** the range of inputs in farm operations, *** and the extent
of fluctuations in prices.****

~ The more refined measures of this type, the inter—quar-
tile and inter—percentile ranges, reflect the same Weakness except in
lesser degree. For example, the value of the quartile distribution is
the same regardless of Whether the items are uniformly distributed be-

' tWeen their quartiles or are concentrated principally at one end of that
range.

The principal measures of dispersion which depend upon
the relation of each individual item to some mean are the average or
mean deviation and the standard deviation of the series. The mean de-
viation of a series is the arithmetic mean of the deviations in the val-
ues of the individual items of the series from the roan of that series.
The deviations are summed without regard to sign. Mills, in his recent
book "The Behavior of P1ices," has used the average deviation as a meas-
ure of geographic variation in prices, ***** and modifications of it as
measures of the monthly****** and yeardto-year variability of
prices.******* Bewloy has also used a modification of the mean devia—
tion to show variations in prices.****$***

 

* Yule, p. 153.
** Minnesota Bulletin 224 — Management Problems of Farmers' Elevators,
p. 11.
*** U. S. D. A . Bulletin 1271 - Farm Organization in Southwestern Minn—
esota, p. 16.

**** Boyle, J. E. — Speculation and the Chicago Board of Trade, p. 122.

***** page 162.

****** page.39. The mean deviation as a percent of the mean.

******* page 49. The mean deviation of the link relatives of the an-

nual average prices.

******** Average of the percentage deviations of individual relatives
from their geometric mean—
"London and Cambridge Economic Service." Special Memorandum,
No. 5, February 1924.

i}

158

The other principal measure computed in terms of the va-
riat ion of each item from a central value is the standard deviation.'
Tho standard deviation is the square root of the arithmetic mean of all
deviations, the deviations being measured from the arithmetic mean of
the observations. The arithmetic mean is chosen since the sum of the
squared deviations from it are less than for any other value of the va-
riable. The standard deviation is probably the most widely used measure
of variation. Yule says it, "should all-rays be used .as the measure of
dispersion, unless there is some very definite reason for preferring
another. measure."* In this respect it is much like the arithmetic _
mean. The great value of the standard "deviation lies in its subsequent
importance in the statistical analysis, .for example, in the problem of
sampling and of correlation. The- standard deviation possesses the ad-
vantages of rigid definition,- ready algebraical treatment, dependence on
all the values of the series, and', in general, of being less affeeted
.~;.-y the fluctuations of sampling than the other measures of dispersion
wl ich We have described.

The common measures for comparing the variation of two
series are ordinarily constructed on the assumption that there is a re-
lation between the size of the items comprising the series and the ab-
solute variatienamong the items of the series. This might be accom-
plished quite directly by computing the goorxetric means of the series
and measuring the deviations by their ratios to the geometric means.
This, however, is not often done. The simpler method is usually to
correct the measures for size by dividing the measures of absolute vari—
ation by some mean. The standard deviation divided by the arithmetic
mean gives what is called "the coefficient of variation," the most ,
widely used comparative measure of variation. It must, however, be in-
terpreted carefully in mny economic series, especially when the series
whose variations it measures contain positive and negative items and
the means of the series are in consequence srraller. This night, for
example, occur in considering. the labor incomes of two groups of farms.

 

* Yule, p. 144.

e.“

‘u

(2) The Measurement of Time Series;*

Movements.

._.._.—~

The distinguishing characteristic of time series is that
they consist of observations with regard to time. Each item in
such a series is the outcome of the continuing operation of the same
forces which caused the preceding ones. It may be partly the outcome
of influences set in operation by earlier items in the same series. ,
At each moment these forces, opera ting in their economic and physical
environment, result in a temporary equilibrium of economic situation.
But. this is only a transitory phase or a crossaseCtion of the contin—
uous sequence of events. It centains within itself elements of growth
or development which may be expected to bring about its own modifica—
tion, or for that matter successive modifications of the future flow
of events. Our analysis of soonomic time series is aimed at the cam—
prehension of this dynamic manner in which econOmic fordes act in the
unrolling of our drdinary sequences of events; Our recorded items re-
present simple measurements at regular intervals of a series of cross—
Sections of the flow of phenomena. We seem unable to grasp or compre—
hend this dynamic flow of the forces themselves. We are able to per»
eeive them only by their manifestations in events. This limits our
analysis to a careful examination of the recorded phenomena, from which
we may be able to draw some inferences as to the nature of the under—
lying forces and the regularity and strong th with which they act.

' f The fluctuations to which any series is sub--
Tynes of i * just are the result of a multitude of influences,

movements* . having effects differing in direction, intensity

and duration in relation to a given series, and
‘ varying in their relative effects between one ser—
ies and another. Moreover, it is impossible to express quantitative-
ly all the factors known to affect a given series. Fortunately, it is
sufficient for most purposes if the effects of only those factors
which exert a uniform, regular, or regularly recurrent influence on
the magnitude of the items in the series can be expressed.

 

* The material in this section was prepared by the following: 'John A.
Hopkins Jr., Iowa State College; 0. F. Clayton. Michigan State College;
C. M. Purves, U. S. Bureau of Agricultural Economics; V. R. Wertz, Uni—
versity of Ohio; H. B. Killough, .Brown University; B. A. Holt, Univer—
SitY'0f_Minnesota. The committee on the handbook fitted the contribu»
tions of these six into one article, thus eliminating overlappings.
Specific credit will be_given for all the major divisions, but it will
not be possible tonnention the author of each separate paragraph.

**The classification here presented is that of Professor Day. This
section was mostly contributed by Hopkins and Clayton.

’.'-" ¢ . ‘ '_ .. _ 7.

igcrc
r-—_.1eo “

I

Time series are subject to threefmain types of influences whose
effects are relatively regular and continuous:

(1) The magnitude of the items of a time-series is affected by
certain underlying, long—time influences, tending to produce in the
series a persistent upward or downward movement extending over a re—
latively long period of time. Professor Wesley c. Mitchell suggests
the following classification of those long-time influences: "(1) Caus—
es related to changes in the number of population, (2). causes related
to the economic efficiency of the population-its age, health, educat—
ion, technical knowledge and equipment, methods of cecperation, meth—
ods of settling conflicts of interest, and many other matters, (3)
causes related to the quantity and quality of the natural resources
exploited by the population. "* . « ~

The second sort of regular movements are the cyclical or re~
current. These occur with a sort of rhythm and more or less period—
161tY. The term cyclical has been applied to a large number of dif—
forent sorts of va.riations, but more careful examination finds it dif~
ficult to demonstrate the true cyclical nature of any very large pro—
portion of business_fluctuations.' Attempts have been made, but as yet
with but little success, to demonstrate the regularly rhythmic influe—
nces capable of causing economic. cycles in the strict sense. Corre—
lations have been sought between economic fluctuat-ions and cycles of
solar radiation or sun spots. Cyclical f1u.ctuations in the mass
psychology of the business man have been suggested but not demonstrat~
ed. ' The length of time required to make. adjustments in the product—
ion of crops or livestock would appear to furnish a sound reason for
a period of rhythmic fluctuations in the production and prices of some
farm products. But we would-hardly expect such fluctuations to con~
tinue to show as great an amplitude from cycle to cycle. That is, we
might expect the resulting adjustments in production to approximate
the normal more closely each time, so that the cycles~arising from
this cause would become of less and less ampl-itude, unless stimulat—
ed by new forces from outside the industry cohcerned.

This does not mean to deny the existence of more or less re—
gularly recurrent fluctuations in price or production data, but to
raise the question whether most fluctuations that‘ have been called
cyclical may not be of the unique. or episodic type mentioned later.
Of those which do seem regularly recurrent, only a few types. suCh
as the seasonal, have been satisfactorily explained.

The third type of fluctuation. is the seasonal. This is in
fact a type of the regularly cyclical sort. In the study of -agri—_
cultural series, the seasonal fluctuations are of great importance.
The recurrence of the growing season is the outstand-ing character-
istic of agricultural business, and the supplies of farm produce by
their conformation to it give us a typical yearhtoayear cycle in
prices also. Here we hays a clear case of a recurrent or rhythmic
cause of fluctuation which results in oscillations from maxima to
minima with regular periodicity.

" * Mitchell, Wesley 6., Wes, 5%

 

 

The distinction between "cycle" and " periodicity" becomes
important here. This distinction was stated by Dr. F. E. Clements
at "A Conference on Cycles" held in 1922, as follows: "It seems de-
sirable to use cycles as the inclusive term for all recurrences that
lend themselves to measurements, and period or periodicity for those
with a definite time interval, recognizing however, that there is no
fixed line between the two." (Cited by Mitchell, Wesley 0., in Bus—
iness Cycles, The Problem and Its Setting, p. 377.)

. g It is well to recognize that the distinction between cyclical
-and secular is not absolute. What may seem like a secular trend may
prove in fact to be one movement in a long cycle.

_ Of the non—cyclical fluctuations, some may be called episodic
_(1). The disturbances in prices or other data resulting from a war
. Lre of this type. Under the same heading may be grouped the abberat-
ions from the introduction of new methods of production in an industry,
or the opening up of a new producing territory, or the uncovering of a
new and different sort of a demand for a product. Episodic movements
are characterized by a deviation from the otherwise normal trends which
is likely to grow more pronounced as the new influence gathers force.
If the new force persists, the old trends may reappear, perhaps some*
what modified, or perhaps much the same as before, but at a different
level from what might have been expected from a projection of the or—
iginal lines of trend. This is illustrated by the disturbances in the
prices of cattle with the opening up of the new range country between
1875 and 1885, and again by the change in trend shortly after 1902 as
improved methods of feeding reduced the costs of beef production and
caused a lowering of level of the-seasonal and secular trends.

If the disturbing influence is of a temporary nature, the old
trends may again appear at essentially the plane of the projected trend
lines. Thus an outbreak of hog cholera would be expected to cause a
decided deviation from the usual levels of price and.production, but
after it was past the old trends would be expected to reappear essent—
ially as before. The outbreak of war is of the same general nature,
except that a disturbance of such magnitude as a maJor war would be ex—
pected to stimulate the adoption of new methods of production result-
ing in a permanent displacement of the trends as mentioned in .the last
paragraph.

The chief characteristics of the episodic fluctuations are
that they are generally unpredicted and unrecurrent. Each case is to _
be regarded as unique and requires a different analytical approach.
They vary widely both in duration and in amplitude and range, from the
disturbance occasioned by a railroad strike lasting a few weeks, to
that caused by a major war lasting several yearsq

Nevertheless episodic fluctuations may sometimes occur with
an appearance of a rough sort of regularity. This may be illustrated
by the fluctuations in the prices of cattle since 1870 in which there

 

(1) See Day, E. E., Statistical Analysis, New York, 1925, pp. 302-306.

s,‘

\s

162

have been at least four outstanding episodic disturbances which occurr-
ed at intervals of from 7 to 17 years if measured from peak to peak. 01‘
from 13 to 18 years if measured from trough to trough*. In each ep- .
isode a different and nonrocurrent influence was to be found as the has»
10 cause. Unless a careful study of each episode is made. such a series
of fluctuations will be mistaken_for a genuinely cylical or regularly
recurrent type of fluctuation. ‘

There is another type of non—cylical fluctuation which may be
called the'irregular or fortuitous. This comprises the fluctuations
that cannot be exPIained by continuous operation of underlying forces
a; in the case of the secular trend, nor by regularly recurrent rhythms
e: in the-case of cyclical or seasonal trends, nor by the intrusion of
powerful new and unusual forces as in the case of the episodic move-
ments. Even when all these movements are explained, the ordinary series
of economic data will show some residual fluctuations for which no re-
gular causes are apparent. Many of the so-calléd irregular fluctuations
are usually discovered, on more thorough analysis. to constitute minor
epiSodic movements. That is, the more careful and thorough the analy-
sis, the fewer unexplained variations are likely to remain. However,
there are always some for which no reasons can be assigned. These may
- represent erratic market information, or erratic reaction on the part
of sellers or buyers of produce to information which may be essentially 2
correct. Or they may represent chance combinations or forces which
give the market a temporary and unexnected turn.

Secular trend has been defined as."the long-
Secular time tendency of the items of the series to grow
' Trend** or decline regardless of the temporary seasonal
and cylical and irregular movements**¥9 .This
. definition emphasizes the fundamental or organic
character of the changes represented by the trend and the “independ—
ence of these changes from those of a seasonal, cylical, and spor-
adic character. If, however. the seasonal movement is causally re»
lated to the cyclical movement and the latter to the trend, then the
empirically.determined magnitudes representing the trend cannot be
said to represent the changes which the series would undergo regard-

 

*VSee Iowa Agri. Exp. Sta Research Bulletin 101.! A Statistical study
of the Prices and Production of Beef Cattle. pp. 349-357.

**Contrihuted mostly by Hopkins and Clayton.
~¥**Falkner, Helen D.. "The.Measurement of Seasonal variation VOILXIX",

June; 1924, p. 167. Where not otherwise stated, the references in
this section are to the Journal of the American Statistical Assn.

less of the other types of f1uctuation.* The trend, in fact is merely an
empirical representation of‘a particular type of change characteristic of
the series over a period of time. Neither the concept of trend nor the me-
thods of its determination warrants the assumption that the magnitudes re—
presenting the trend are independent of seasonal and cyclical influences.
The statistical procedure involved in the isolation of the trend and the
seasonal and cyclical fluctuations of a series, serves to exhibit the na—
ture or character of the changes, but in itself tells nothing as to the
relationship between the changes themselves or between or among their un-
derlying causes, The search for causal relationships lies beyond the pro-
cedure for determining and isolating the characteristic fluctuations of the
series. Perhaps by analogy it might be said that the_prob1ems are primari—
ly those of anatomy rather than those of physiology, that the relationships
established are structural rather than-functional in Character.

Logically, the problem of the secular trend is the problem of
determining a_value which most nearly represents the "normal" condition for
the data at each unit of time throughout the series. This normal may not
be the average trend, as normal means that condition which tends to prevail
over a long period of time, while an average condition is the result obtain—
ed from combining a number of actual values. ' - '

The methods used in determining trend.may be described under
three heads: (1) free—hand curves, (2) moving averages or medians, and (3)
mathematical lines or curves. A few of the more common of the mathematical
methods of fitting lines or curves are the method of least squares, applied
either to the natural date or to the logarithms or reciprocals of the data,
the method of moments, parabolic equations, and hyperbolic equations. Each
of the above methods of measuring trend as well as others less commonly
used are applicable to certain kinds of data. Experience in curve fitting,
which familiarizes one with the shape of each different type of curve, is
essential before the best curve for any given time series can be readily de-
cided upon. One must also have a thorough understanding of the data being
worked with.4 Oftentimes a trend may be fitted to data which appears entire-
ly reasonable, but a knowledge of the data will reveal that certain factors
were at work during one or more periods which caused the data to be abnor-
mal. Taking these abnormalities into account when fitting the trend may re‘
sult in entirely different results. The first step in the fitting of a
curve to data is to plot the data in order to obtain a general view of the
fluctuations from period to period and the general trend throughout the en-
tire series. This examination alone will often tell the type of trend that

 

* Cf. FTenk,_Lawrence K., "Long Time Price‘Trends" Vol. XVlll, Sept., 1923.
pp. 904—908, in which the interesting hypothesis is advanced that "the long
term or secular trend in economic activities is generated by cyclical fluct—
uations". And in a footnote to the same article, Mr. Frank observes that
"it is an interesting question whether the cyclical fluctuations which gen-
erate the secular trends are not themselves genarated by'seasonal fluctuat-
ions". See, also, Vol. XX, Dec.1925, pp. 543—545, Smith, Bradford 3.,
"Error in Eliminating Secular Trend and Seasonal Variation before-Correlat-
ing Time Series."

‘1

_ \

164

is present. But not infrequently it will be necessary to verify this con-
clusion before proceeding further.

In free—hbnd fitting care should be tench ..

EESEZEEEQ In free—hand fitting care should be taken
fittin - to see first that.the line drawn describes a

straight line or a smooth curve, and second, that

it leaves equal areas above and below the trend

line: that is, that it passes through the center
of the points plotted and conforms to the direction of drift of the whole
series. If a straight line is fitted in this way, it should pass through
the mean of the whole series at its center.and the mean of the two end-sec-
tions of the series. The use of free—hand curves is objectionable on the
ground that the method lacks precision. The results obtained depend on the
individual, not on the method. However, as used.with certain checks de-
scribed later such trend—lines have their value also. Moreover, as will
be pointed out presently, the "appropriateness" of any curve obtained by
the rigid application of mathematical procedure must in the last analysis
rest upon the Judgment- of the investigator. Dr. W. I. King even suggest3
that "the eye is the ultimate court of appeal as to the correctness of the
locati.on of the trend". * The advantase to be gained by the employment of
objective rather than free-hand methods are therefore continent upon a jud-
. icious choice of formula. Such a choice must rest upon an understanding of
the characteristics of the line or curve employed and the assumptions under—
lying the method of fitting the line or curve to the data.

A moving average may be used either to elimi-

’ Moving. ' ‘ nate regular cyclical fluctuations or to smooth

averages. ' out random variations. If a cyclical fluctuation

4 is evident, the moving average should be of the

length of the cycle.** Such an average will not

extend the whole length of the series, but will be short by one item at
each end in case of a three—year average, or two at each end in case of a
five-year average. Its advantage is in its ease and simplicity of computat-
ion and its flexibility. In case of chanPes in the rate or the direction
.of the trend, the moving average will follow the new trend whereas a math—
ematically fitted line would fail to conform to the change in direction,
and a new trend would need to be computed from the point of inflection.
This is a somewhat questionable advantage, however, since one never knows
whether the change in direction shown by the moving average in such case is
merely temporary or has come to stay for a while. A moving average in it-
self is no evidence on this point. Episodic disturbances upset moving av-
erages severely.

 

‘* King, Wilford 1., "Principles Underlying the Isolation of Cycles and
Trends", Vol. x1x, December, 1904, p. 471..

*fFor an excellent discussion of the characteristics of the moving
average as a device for representing trend, see Mills, Frederick C.,
"Statistical Methods Applied to Economics and Business", pp. 260-271.

. 165

TABLE I. MOVING AVERAGES OF HOG PRICES A9 CHICAGO ADCORDING T0
FOUR DIFFERENT PERIODS.

Original: 3—year : 4-year : 5-year 7-year

 

YEar :1 Data ': Moving Avg.: Moving.Avg. Moving Avg. 1 Moving Av.
1927 : 10.10 : : : :

1926 :‘ 11.95 : 11.25 - z , : - :

1925 : 11.70 : 10.63 : . 7.42 : 9.89 : .
1934 : 8.25 : 9.13 : 9.48, : 9.68 z 9.55
1923 : 7.45 _: 8.25 : 8.70 : 8.96 : 10.09
1922 : 9.05 : 8.28 : 8.98 : 9.59 : 10.91
1921 : 8.35 : 10.42 : 10.96 : 11.28 : 11.74
1920 : 13.85 : 13.30 : .13.3O : 13.29 : 12.73
1919 : 17.70 : 16.35 : 15.21 : 14.52 :‘ 13.04
1918 : 17.50 : 16.80 : 15.54 : 14.78 : 12.75
1917 : 15.20 : 14.12 : 13.68 : 13.41 : 12.73
1916 : 9.65 : 10.62 : 11.18_ : » 11.51 : 11.92
1915 : 7.00 : 8.28 : 9.14 z 9.65 : 10.47
1914 : 8.20- : 7.80 : -8.00 z -8.12 a 8.92
1913 : 8.20 : 17.98 : 7.70 : 7.52 : 8.02
1912 : 7.55 : 7.47 : 7.74 : 7.90 : 7.71
1911 : 6.65 : 7.70 : 7.74 : 7.75 : 7.53
1910 : 8.90, : 7.67 : 7.42 : 7.26 : 7.22
1909 : 7.45 : 7.37 : 6.37 : 6.96 : 6.94
1908 : 5.75 : 6.42 :, 6.02 : 6.88 : 6.61
1907 : 6.05 : 6.02 : 6.17 : 6.15 : 6.40
1906 : 6.25 : 5.85 : 5.76 : 5.69 : 5.99
1905 : 5.25 : 5.55 g 5.67 : 5.74 z 5.91
1904 2 5.15 : 5.47 :_ 5.75 : ~ 5.92 3 5.94
1903 : 6.00 : 6.03 : 5.92 : 5.85 : 5.79
1902 : 6.95 : 6.28 : 5.99 : 5.81 : 5.48
1901 : 5.90 : 5.97 : 5.74 : 5.59 : 5.28
1900 : 5.05 : 5.00 : ' 5. 10 : 5. 16 : 5.06
1899 : 4.05 : 4.32 .: 4.43 : 4.50 : 4.69
1898 : _3585 : _3.85 : 3.95 : 4.00 :

1897 : 3.65 : 3.63 : 9 : :

1896 : 3.40 : : : :

 

Table 1, showing moving averages of yearly hog prices at
Chicago computed according to four different periods, reveals some
of the characteristics of this type of trend measurement as applied
to a typical economic series. - It is apparent that the cyclical
movements in this case are of an irregular length and amplitude,ae
is typical of most economic series. If measured from trough to
trough, the cycle before 1914'varies from 3 to 5 years. If measured
from peak to peak, each is four years in length. ‘After 1914, the
outstanding episodic movement caused by the European war reaches its
peak in 1919 and the trough of the subseguent reaction was reached
in 1921. Had the cycles been of uniform length and amplitude, we
would expect a straight line to result from a moving average with a

['4'

15's

period equal to the length of the cycle. Had the length been uniform
and the amplitude uneven, we would expect a Series of cycles of smaller
amplitude to appear. What we get instead is a moving average in which
the cycles are still evident, but less pronounced than in the raw data.
The five-year moving ayerage differs from the three-year one only in
being samewhat smoother. _ Since four years seems to be the nearest the
usual length of the cycles, a four-year moving average was computed, as
shown in Column 5, and this was centered by averaging each adJecent pair
0f averages, The seven-year'average.;which'is practically equal to the
len8th of two cyles; most‘nearly removes the'cyclical influences. Fig—
ure 1 shows the effect of the different movihg averages.

.- In the period after 1914, With two dominant movements of an
edisodic nature and-of unequal length,'no one of the moving averages
cowputed results in~a line of trend which seems particularly signifi-
cant. -For these periods, the irregular fluctuations were simply
smoothed out more or less, depending on the.length of the period cov—
ered bY-the average-in question. When the price turned'up sharply
after 1914, it will‘be observed‘that the moving'averages, and par—
ticularly the one with the longer period, gave a line each point of
which is above the actual data for three years and then below for
three years.‘ When the price turned down sharply after 1919. the
moving average gave*a line which was again above the actual prices.
This means that in both of these cases a change appeared in the trend
line long before it really happened. ‘The sharp break upward.0f prices
in 1916'began to show in the moving average 3 years earlier in the
7-year moving average, and 2 years earlier in the 5eyear moving average.
When the-break downward came in'1920, the high prices of the war years
remained in the moving average and kept it from reaching_bottom until
equally too late. All moving'average series of prices should stop
with 1915 and resume about 1925; To continue them.during the dis-
turbed period of 1915.1925 adds'only confusion. Anlexcellent example
of this-confusion appears in Figure 2, reproduced from Chart B of the
report of the Business Men's Cowmission, in which the use of an-ll—year
moving average of prices of farm products make agriculture start down-
ward in 1914; 5 years before it happened. But it does not take an
episodic disturbance such'as that of the World war to make moving av-
erages confuse the picturew. Any time that a series is concave up-'
ward, the moving average will run ahead of the data, and it will run
behind when the series is convex downward. It will be obvious that
deviations measured from moving averages are not very dependable when-
ever disturbances occur or convexity or concavity appear in the data.
In particular, the exact point in time at'which the actual series
crosses the trend line will have little significance. A chart such
as Figure 3, reproduced from page 344 of the July 1927 Journal of '
Farm'Economics,.in which a comparison of deviations'from the above
described 11-year moving average,with another 11-year moving average
is made the basis of conclusions as to the relation between agricul-
ture and business, the essence of which is the time when the devia—
tions occur, must therefore be scrutinized very carefully for falsie
fication produced in connection with concavity, convexity and episodic
movements. Using a long period spreads these errors more evenly,
but over more years.

\\‘

167

 

 

 

 

 

 

 

 

 

 

3.0K .049 ow_m< .29 Q? .89 . 89 , .BW
3% 33%? win 8N
litl... XT 3330: .v‘kmw
|.I.....|..t NSVQ $195.0? ..
inewfi 3%
om.
\/ . \xV!“\\\/MW: 09%
\\\/J&/.& .. \\ /
71.. .. . _ \ .... ‘ A -83
. ‘ . . , of
_ 9E
\
/ \\\
. \ 3.3
w

 

 

 

 

 

 

 

Q .,

.07...

 

 

 

39mm 60: no mmwémzq 02302

/68

0mm. m

mm.

ONE

9m. 9m. mom. 00m.

mmw.

0mm.

mwfi

owm.

fi—~_

 

4..

u..—

_.._ ._‘. «.u.___._fil;.

_—_.

.fiau

 

.—~_m

 

 

 

 

 

 

 

 

 

 

 

 
 

mmb 50». T \wQ‘xoz \w
«09?? ax .

 

 

,1
K

 

 

 

 

 

 

 

 

 

 

 

 

.m‘mfixéx $.SfiSusu \Ok3§39.tmuc.b>‘

N

.0;

 

 

 

 

 

. 3 $3me .833on kink $0 036%

 

 

 

 

 

 

 

QQ\H “im‘ n 0‘2

 

 

 

mum.

Ow

on

GM

ON.

thu mun»

mmiimm J<EDFJDBKO<ZOZ OF FUDOOEQ 2E<u n5 Orr/xi

.0... \\ ‘. .hpwnommonm mmofimdn an moowum Hausfisghww Boa 535938 .33 mnowmaofimov mmocfimso. hp
2 @2833 $3.5 Hahdfisghmahmg hflgwpmaon 3.33 .hfipwpom mmonwmdp Hgoqow 5 msg#03353 03939
_ _ 1800 an dawnmmaooom noon 9%: $0an oawmmfiofip Hmnsfisghmw mo nosom mammmnonsm $332 on» 5 no

.30ng amusfldowumddon on 33.3 35.08% and.“ no 39.2 93 5. 305.3853 wEwpupnonm 93. .n .mwm

(«WE mam. ow?

.

 

 

 

 

   

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ME. 022 .009 009 mmfl. . 0mm“ 39 Own. 059
.HJ111 u-J ~4..q44.—_._—-—.J~«_1fi~qq«~4~. ‘#4<1 .._. 1-.“
‘ V m . M _ ,
. . 4+3}ng \CSFuxmmmzfiam go Xmaz. mochfiu. W cm
W
om
‘ . 09
g
o:
, on.
cm
on
. Iyfivmrt‘a...‘ . ._ vu=u~w _ ~ . . u \ :.. . .. ‘ . ‘ .:.
wirvobunrhufltm \ ... ..... \«fhdkmw . .. . . .. . ,w . .. . . . : .. . . . . . . . .. . 8‘
. o:
,
um...» M . a . L ON“
. . oor..>¢02_>02 .Er: .3ny gashouagzoz
, N M _ 9533mm :mE mo 0E»:
, . - 1. M _ _ _ _ kzmummu .

mmdx/U mmuzaam azq 3,65 mo_aaq<mzbvo&¢<

 

II .o

’5'} T

170

In the matter of the length of the period‘in such a case, :
Mills writes as follows:“ - - - a moving average should, in general, 2
be based upon a period at least equal to the period of the cycle, ‘
and preferably equal to some higher multiple of that period when the'
data are at all irregular...... But when the underlying trend de-
parts materially from the linear form, ...... the error involved in
the use of any moving average increases as the period of the average
increases. If a moving average is used in such a case to measure
the trend, the period of the average should be the shortest which
will serve to average out the cycles: equal, that is, to the average
length of one cycle."*

Moving ,' > Since the aim is to fit a line which shows:
median. the dominant tendency of the series, the' ‘

moving average has the obvious disadvantage

that it is subject to the effect of wisely~ .

deviating items in the series. %s suggests..
the use of a moving median. If used in a series having??% uctuations,
however, the course of the moving median is likely to be erratic. More-
over, if the seasonal variation of the series has a marked upward or
downward tendency, this will give a corresponding bias to the median
values obtained. Seasonal variation should be eliminated in such
cases before computing the moving media.** ‘

Fitting b1 ' i It has already been observed that the use of
eguations. f a mathematical equation involves an assump—

tion as to the fundamental tendency of thev
series. A choice must be made at the out:
set as to the type of curve appropriate for
the given series. This choice should be made in view of what seems
to be the nature of the phenomen This does not mean that the an—
alysis should be carried through a pro—supposition as to the conclu-..
sions. But it should be recognized that the choice of one out of
two or more pOSSible methods of analysis itself assume the nature of :3
the phenomena to be the type to which the chosen method is most suit-1-
able. In studying the secular trend of a series of prices, it is
poSsible to compute a straight-line trend, or to use any of a large'
number of culvilinear formulae. Exactly the same body of data is
capable of yielding any one of a considerable number of lines of
trend, depending upon the formula used. ' The formulas are useful
in determining the rates of change and in describing the series
after the system according to which it is varying has been at least
tentatively decided. The computation of several forms of trend
will help to discover what that. system actually is. But a more
rationalistic study Of. the data will be the starting point.

The formulas finally decided on will be used in des—
cribing the data in quantitative terms to show what it is and how

 

* loc. cit. p. 267.
**Cf. King, Willford 1., Ice. cit. p. 474.

l7!

it fluctuates. The explanation of why it behaves as it does must
depend to a large degree on the qualitative considerations added to
these quantitative ones.

A18°. as Dr. W. I..King observes: "In most instances there
is no way of knowing in advance whether trends are segments of straight
lines, of curves representing waves of great length, or of mathematical
curves of some other type".* It follows that trends determined for
short periods can only be tentative and subject to correction as data
for longer periods serve to exhibit the underlying long-time tendency
of the series. ‘1 further consideration is that the fundamental tend-
ency of a time series is itself subject to change, so that the math-
ematical law which properly expresses the fundamental tendency of a
series for one period may have little or no application to subsequent
periods. There is no secure basis for assuming, for example, a
straight-line trend for a time series in the future, simply because
such a trend has been characteristic of the past. If such an assump-
tion is made, it must rest on consideration of an economic sort, not
on the general validity of the mathematical law used to characterize
the "growth"of the series.

Strai t-line . ‘ If the series is changing upqard or downward
EEEEQE: ' in a straight line or moving horizontally,

an approximation of the trend can be obtained

by passing a thread or ruler through the

series-as plotted to a natural scale. ,But
such a trend will of course be only roughly approximate. If any

'further analysis of the data is to be made, a mathematically deter—

mined straight line is needed.

- If a straight line is to be determined mathematically. some
care should be exercised in the selection of the period to which it
is fitted. In the first place, the line should be applied to one
homegeneous body of data only. Thus in determining a significant
trend of hog prices we could not compute a straight line through the
entire period from 1896 to 1927 because it would be distorted by the
widely divergent data of the war and post—war period. If it were
desired to compute a trend for the post-war period, it should be
based only on data after the more temponary and violent influences
of the war had subsided.

Another consideration is that the direction of the trend
line is more or less dependent on the phases of any cyclical or ep—
isodic movements with which the included data begin and end. Thus
if the period studied were to begin with the low phase of a cycle
and end with the peak of another cycle, the trend line of an as-
cending series would be more steeply pitched than it should. The
trend line should begin and end in corresponding phases in order
to be truhy representative of the period included.

 

* 10c. cit. p. 469.

172

The formula for the straight line is y s autbx where (y)
is an ordinate or point on the line of trend, (a) is the point of
origin of the line, (b) is the amount by Which the trend line rises
in each period of time, and (x) is the number of intervals between
the point of origin and the h t of time in question. The problem
then is to find the values ofa ') and (b) which will determine the
straight line most closely approximating the trend of the data given.

Method of The most usual method of fitting a straight
Least line to a curve is that of least squares.*
S uares For certain types of data it is the best

vmeasure of trend; but it is probably more

often misused than any other method because
of its ease of calculation, precision, and simplicity. It should be
used only for data with fairly uniform.vnriations_and preferably only
when analyzing a large number of observations: Should the variations
from trend be very irregular in a small number of observations and one
or two unusual variations in the same direction occur near the extreme
end of the series, these variations have an extreme influence in de—
termining trend because the square of the deviation from the trend is
so large that it unduly weights that item, and the trend line this
fitted does not represent normal conditions. It is a matter of com—
mon knowledge that a shift in the interval for which the calculation
is made may give significantly different trends when the lines are
fitted to time series by the method of least squares. It may easily
be argued, therefore, that the system of weighting involved in the
use of the least-squares method is devoid of economic significance
when anplied to time series. Dr. Ingraham points out that slight
changes in the group of years included produce marked difference in
the increments shown by the least square trends.** And he suggests
that the discrepancies are due, for the most part, to variations in
the amplitude of the flucations of the original series, but also in
part to Special characteristics of the least squares formula. In
the light of these considerations, Dr. Ingraham suggests that " a
line of coordinated means ... calculated from a three—year moving
average"*** avoids the instability of the least squares line, and
hence is more satisfactory for determining the secular trend.

 

* For variousstylesof equations and their straight-line forms,
see Huntington, E. V., "Curve—Fitting By the Method of Least
Squares and the Method of Moments", in Handbook of Mathema-
tical Statistics, by H. L. Rietz, Editor-in—chief. p. 63;
a180 any of the newer textbooks in Statistics.

** Cf. e.g.. Ingraham, Olin, "The Refinement of Time Series",
June 1925, Vol. XX pp. 231-233.

Hot Ice. cit., p. 233.

t4
«2
C):

Mr. O. Gressens* has pointed out, however, that the pro—
per employment of the least squares method does not admit of an in—
discriminate choice of the period to be included. In particular,
as Mr. Cressens suggests, attention should be given to the follow—
ing considerations:, (a) The concept of a secular trend implies
that lines fitted for short periods are necessarily tentative and
to be regarded only as approximations, subject to correction when

subsequent data more clearly record the fundamental tendency of
the series,‘ " »

(b) A line correctly representing the trend of a series '
for a given period is not necessarily yalid,forhcthereperiod8. since
the fundamental tendency of the series is subject to chang6..

(c) The least squares method gives heavier weight to those
items close to the terminal portions of the series. Hence. an 68-
pecial bias is introduced in fitting the line unless care is taken
to see that each and period is in the same phase of the cycle.

A consideration of the statistical propzéfties of the least
squares method leads Professor W: L. Cfum "to sug 'st deviCES"by which
more Complete knowledge may be gained of the actual significance of
the leaSt squares operation in any particular case."** After call—
ing attention to the fact that the annual increment obtained by the
method of least squares is in effect a weighted average of'ratios,

'Professor Crum proceeds to test the precision of the least Squares

method by studying the form of the frequency distribution of the
ratios about their mean. He concludes "that the method is not
automatically safeguarded. In a number of the Specific instances
studied, the process seems to yield results which are satisfactorily
precise. In other cases it falls far short. No general rule

stands forth: It seems necessary to examine each case with care."***
As a test of the precision of the results to be eXpected from the use
of the least squares method in a Specific case, Professor Crum sugs
gests the construction of a frequency diagram of the ratios as a step
in the computation of the trend. -

An application of the different methods of measuring
straight line trends to actual data may bring out more clearly some
of the problems involved in measuring trend. The data used in Fig.4
are an average of the monthly average price of No. 3 yellow corn at
Chicago, from November to May from Nov. 1899 to May 1927. deflated
by the Bureau of Labor Statistics allwcommodities index number
(19133100). The plotting of the data shows that there is a sharp
break in prices between 1919 and 1920, and that the lower price
level has continued since 1920, which makes it necessary to divide
the data into two series in order to determine the normal price
trend during the entire period.

 

* loc. cit. pp.552—553. See also, in am issue, pp.565—568, the
report of a "Meeting on Ways of Using Recorded Data to Estimate
Future Trends." ' '

** Crum W. L., "The Least Squares Criterion-of Trend Lines," June,

1925 V01. XX . 213.
*** loc.'cit. p. 225. e

‘n,

a
‘.
\\

W/

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dieflczyed Mu ;o 1140' co‘rq pHces'Na3fyd/gw dingo ”3.5.9 #1941
, cams --~~-- Hirer/by mar/70d o leasr aqua/es.
pen '———-fi7fed4y memod of averages
“MEL . ufi’fieo’ by /easr squares, 00/
[9/6, /3/7 and ma (am/fled.
3° FIG.4 ' r:
8S"
60 ~—- + W '
/. L / /f7..-/"\/ V .. .1
4O -
1 I; l l _ | I I l L
[837 99- 0! 06 . 05'. O7 09 lo I 7 29
MlLL-OJ FIG. 5- ‘ I
BUSHELS '
3200
3009 A A -‘
2800 \\ . 'AAJ:%
. a 'I‘. " /‘r
1600 f4 ‘
‘ KKK
2.400 :/‘:;/./
L
2100‘ /4 ‘ 7'
l V ’ ‘ gun—— Cor-n produqu plus carryover 4&9} ~03}
00/ '-«---—;/= /3567+83.-877x mmxl
1° _: - . ;‘--—-~Log}’-* ssoaw ran/4649 /05/X
'V 1 ~ 1-
' 11.11““. ill'L1‘-t‘_‘a' z
1897 53 Di 03 05 07 09 n :3 15 'n :9 as 23 25 27 29

 

 

 

 

175 f,-'

First a straight -line trend was fitted to each series by
the method of least squares, assuming no knowledge of the causes for
p the unusual variations which occurred from year to year. The results
are represented by the solid lines shown in Fig. 4. It will be noted
that the unusual year 1901 is counterbalanced by the low prices of
1899, 1905 and 1906. But the unusually high year 1917 has no low
years to offset it, which gives the line too steep a slope. In the
shoit series from 1920 to 1936, the unusually low prices of 1921 and
1932 and the very high price in 1924 make the line of least squares
indicate that the normal trend of prices was upward at the rate of
three cents a year, which makes the normal price of corn to be ex—
pected in.1930, when adjusted to the present price level, to be 106
cents per bushel. to

A knowledge of the facts, however, shows that the price
level from 1916 to 1918 was advancing at a very rapid rate, and as ‘
is usual in a period of rapidly advancing prices, the price of raw
products such as corn advanced first. Even if deflated by the all—
commodities index, corn prices were higher than they actually would
have been in a normal period. It was therefore illogical to‘com-
pare the deflated corn prices in 1916 to 1918 with earlier periods
because they were not sufficiently deflated. . .

Regarding the period since 1920. it must be stated that
Where data fluctuate as widely as in this period, a longer series is
needed to fit a trend by the-method of least squares. One of the
reasons for the wide fluctuation in prices during this period is the
great variation in the production of corn. ' The crops of 1920 and
1931 were the largest on record while the crop of 1924 was the small- .
est crop produced since the unusually small crop of 1901.

With this knowledge of the data available, it yes certain
that the line of least squares when fitted to all observations in
either series is not an ac urate-measure of the normal trend of
prices. -It therefore seemed logical to omit the years 1916 to 1918
from the series of observations,-and since 1901 was a very abnormal
year e end of the series, and would tend to raise the lower end of
the curve to abnormally high values, it too was omitted; and a line
of least squares Ias then fitted to the remaining observations. The
results of this line are shown by the dotted line in Fig. 4. It is
readily apparent that this line of least squares is a better measure
of trend for the years 1899 e 1919 than;the first line used. For
the period since 1920, the best that could be done was to assume that
the normal annual increase in corn prices from 1920 to 1926 was the
same as in the period from 1699 to 1919, and use the average of the
seven years as determining its general level.

The ratio of price to trend was then correlated with the
ratio of supply to trend. This gave a very marked correlation, but
the downward trend in residuals indicated that the trend line was
still too steep for the data.

176

The next method tried was to average the first nine years,
excluding 1901, and the last nine years, excluding 1917, and center
the results on the middle year. This method gave the broken line
shown in Fig. 4. Correlating the ratio cf price to this new trend
with the ratio of supply to it resulted in a materially higher de-
gree of correlation and a more even distribution of residuals. Hence

the broken line was considered the best measure of the normal trend
of price.

MEEEE§_2£ ‘ A simpler method of fitting a straight-1ine
MEEEEEE‘ - trend is the method of moments, illustrated

by Table II."l The equation for the straight
line of best fit for these data is y.6.01+.253x.
This line may be plotted by taking the point of
origin in 1905 as one point and computing the
value of any other point to establish the straight line. Values for
other years than 1905 may be found by adding the value of (b). Which
is .255, successively, or substracting it successively, from the value

of the point of origin. The trend values so obtained are shown in
column 6. ' '

 

Table 11. Computation of Straight Line Trend - by Method of Moments -
Hog Prices - 1896 - 1914.

 

 

 

1 z 2 : 3 : 4 z 5 : 6
:Original : Deviation : Deviation: Deviation : Ordinates of
:’ Items :from Origin: times Orig: s ugred : trend
Year : (y) : (x ) :item(yx) : %x.\ :
: : : 1 ~ :
189.6 : t 3.40 z _9 z . -30.50 2 81 : $ 3.73
1897 : 3.65 : -e : -29.2oé 64 : 3.99
1898 : 3.85 : -7 : ~26.95 : 49 : 4.24
1899 - : 4.05 : ~6 : ~24.30 i 36 z 4.49
1900 : 5.05 : -5 : -25.25 i 25 : 4.74
1901 2 5.90 : -4 : -28.60 : 16 : 5.00
1902 : 6.95 : —3 - : ~20.85 : 9 : 5.25
1903 : 6.00 : -2 : ~12.00 : 4 : 5.50
1904 : 5.15 : «1 r : - 5 15 : - 1 : 5.76
z : : (-197.905: :
1905 : 5.25 z 0 3 o : 0 :. 6.01
1906 : 6.25 x 1 z 6.25 : 1 : 6.26
1907 : 6.05 : 2 : 12.10 : 4 : 6.52
1908 : 5.75 z 5 : 17.25 : 9 : 6.77
1909 : 7.45 : 4 : 29.80 : '16 : 7.02
1910 : 8.90" : 5 : 44.50 : 25 : 7.28
1911 : 6.65 : 6 : 39.90 : 36 : 7.58
s 1912 : 7.55 : 7 : 52.85 : 49 2 7.78 ~
1913 : 8.20 : 8" : 65.60 : 64 : 8.03
1914 : ,‘8.2O : 9 z 73 80 : 81 : 8.29
: ‘- : : Z 342. 055: :
SUM; £x=fllig~5r ixy: I‘M. l5 {XE 57° éXy; bfix‘
Zy:na b‘fifl
a :31 ' (X‘

n
.4. 1‘ 5, .1": 4970

.x’/ 177

QETVES * With absence of known influrences indicating
curvilinear trend, it is usually safer to
assume a straight—line projection than a cur-
vilinear one. But at other times the use of
matlematical curves is justified. Curves fit:

t9? by mathematics are nominate only when the constants used in fitt-

1“ thsse curves have some ielaticnship to the data. When a gun is

first under uniform conditions, the trend of the bullet can be traced

with considerable accuracy by a mathematical formula because the effect
of a certain slope of the gun and the pull of gravity are constant and
the results of these factors can be measured: But in the ever-db nging
economic world it is impossible to determine constant factors which can.
be deed in a mathematical formula and measure what the trend of data
has been and what it will be in the-future. To be sure, constants can
be obtained by applying a series of observations to a mathematical for—
mula, but these constants are not what caused the trend of the data to
be what it was. In fitting a mathematical curve to economic data, one
is not trying to determine the laws underlying the observed quantities,
but rather to obtain a formula by which the trend of the data may be

approximated. .

 

While mathematical curves are somewhat elastic according to
the size of the constants in the formula, they have a definite gen-
eral shape for each type of formula, which cannot be overcome by the
data used: with the result that the-curve often departs.at some point
or points throughout the data from the most probable normal trend of
the data.** .

Series of economic data are often found to conform to the
exponential or compound interest type of curve. That is, the curve
when plotted on a natural scale becomes steeper but at a constant
rate, (or falls with a slope that becomes constantly less and less).
This type may be identified by plotting the series on semi-logarithmic
paper. If the series is of this form, it will aesume a straight line
on the semi-log scale. Plotting the logarithms of the original data
to a natural scale will give the same result. The formula for the
straight line now becomes: log y a a+bx, and the curve may be com—
puted by the usual straight-line method merely by using the logs of
the (y) ordinates instead of the (y) values in their natural form.

This is illustrated in Table III, in which an exponential
curve is fitted to the prices of land in Story county, Iowa, for the
period from 1904 to 1914. The sum of the logs of (y) is found to
be 22.15220. The number of years is ll. . Therefore the log of the
ordinate of trend at the point of origin is 2.01384, or in actual
numbers 103.38. The sum of the (x) items multiplied by the logs of

 

* Contributed mostly by Purves and Hefkins.

** See Mills pp.280—306, for discussion of methods of fitting the
various mathematical curves. .‘Also particularly see pp.300-302
for an excellent statement of tests for the use of the different
curves.

T‘ 178

the (y)'e is 3.86677, and the‘sum of (x2) is 110. Therefore the
log of (b) is the log. 03515 x. The formula thus becomes log

y - 2.01384 4-.03515x. Reducing it to natural numbers, y 3 (103.38)
(1.0841)x. It is obvious that a straight line would not fit these
data, and that a distinctly curvilinear influence is at work in them.

The compound interest or exponential trend gives a reasonably good
fit.

Table III.- Log Curve Fitted to Price of Land in Story
. County, Iowa. 1904-1914.

 

Computed (b)

 

 

(a) _ V 2 Ordinates
Year Y ‘Log Y ‘X ' X.Log Y X of Trend
1904 $72.27 1.85896 . -5 ~9.29480 25 $68.88
1905 74.18 1.87029 2 ' 44 ' -7.48116 16 74.69
1906 - 81.75 1.91249 ‘—3 —5.73747 9 80.98
1907 86.10 1.93500 ' -2 «3.87000 4 87.81
1908 90.95 1.95880 -1 —1.95880 1 95.21
1909 104.15 2.01766 -0 —28.34223 103.38
1910 109.13 2.03800 +1 2.03800 1 111.94
1911 122.55 2.08831 .13 4.17662 4 121.38
1912 128.13 2.107661 +3 6.32298 9 131.61
1913 142.44 "2.15376 +4 8.61500 16 142.70
1914 162.64 2.21128 *5 '11.05640 25 154.74
+32.20900

22.15220 >r3.86677 110

Log y - (inf-bx .

a. 3 (Log 2) b I X.I.IO

n x

= 22215220 3 3.86677
11 110
= 2.01384 (in logs) a .03515

Log y - 2.01384 + .03515x

 

One would naturally be cautious about projecting a trend line rising
in a curvilinear manner Very far into the future. But experience has
justified it within limits repeatedly. A genuine exponential trend
fl' ‘ . *>—,‘- "3.. ~ 1" , .. - ~-~ ' .'

179

msy be present as a result of some influence causing a growth at a
constant rate. If this is the case, the series will continue to
follow the compound interest or log-curve until something happens
to change the rate of growth. I -

' Other types of mathematical curves, and the difficulties
of fitting them, are shown by the following illustrations, based
on the.unrevised final estimate of corn production from 1897-1927,
(plus the carryover of old corn on farms November 1. (See Figure 5).
'An examination of the data after plotting shows that corn production
increased quite rapidly during the first part of the period, and con-
tinuod at a decreasing_rate until 1921, and then decreased materially.
Acreage, however, has not increased since 1912 and has remained fair—
ly constant since that time with a slight'decline from 1924 to 1927.
As this decline has been largely due to increasing.cotton acreage,
decreasing cattle and hog production, and low prices of corn and un-
favorable weather conditions at planting time, it is likely to be more
or less temporary and therefore does not Justify a downward turn in
the longbtime trend of production. The unusually large supplies from
1920-1923 were due to high yields and‘the larger carryover of old corn
”in those years and-were also temporary. These circumstances suggested
that a curve increasing at a decreasing rate until 1912 and then remain-
ing at about a constant level would be the best measure of the normal
trend of corn production from 1897 to 1927. First, two curves of the
hyperbolic type were fitted. The solid line in Figure 5 is the result
of applying the formula' " "

. -, Y- : axb ,
to the data, and the broken line the results of applying the eduation

Y = _____l_,__ o
_ . a-+-bx; - ,

The first curve advances too rapidly for the firet'few years and cons
tinues-too strongly in the later years. It nowhere fits the data.
The second curve is a much-better measure of trend during the last
twenty years, but has too steep a slope during the first five years
and is too high during the second five years. When these.curves are
extended into the future, the difference between them continues to
increase, preducing ‘wider and wider difference in the conclusions
which would be drawn from them. '

“_The parabolic type of curve, Yr- a_+ bx - cxz , is shown
by the dotted'line in Figure 5. This type of curve_tends to be
U-shaped. After the value of X increases to the point where the
ratio of b to c equals the ratio of X to X2, a change in direction
occurs which, in data having a normal trend similar to these, makes
the curve leave the data, 'Increasing the constants or raising them
to higher powers would probably give a curve that'would still more
accurately measure the normal trend-of corn supplies. It should be

180

clear by this time. however, that the constants used in the curves
are not measures of factors which determine the trend of the data,
but rather the results of fitting several observations to the for-
mula. . '

A generally satisfactory test of the fit of the two curves
may be obtained by comparing the standard deviations of the differ-
ences between the points on the curves and the original data. The _
closeness of the line of -trend to the points which were to be fitted
cannot, however be taken as the sole consideration. Conceivably it
may be possible to compute a curve which will pass directly through
each of the points concerned. Yet Such a curve would be of little
or no use. lhat is desired is a smooth curve which will sweep
through the points plotted and will indicate as Simply as possible

the general- trend of the series. The greater the simplicity of the " L

formula and the easier the computations involved the -better. The
curve should be explainable in terms of the forces which seem to be
causing the fluctuations in the series concerned. It may well be
more or less a coincidence that the nonnal trend of the data and a
mathematical curve follow the same path, not due to established re—'
lationships, but to accident. It does not follow that this coin—
cidence is going to continue. ' “' ” ‘ ‘

.  When a second degree parabola was fitted to the data in
Table III on land values in Story county, Iowa, it gave a decidedly
better fit than the exponential curve. The standard deviation of '
the_differences from the parabolic,curve.Was $2.47; from the expon-
ential curve $3.27. It is quite possible that a third-degree par-'
abola or some other form of curve would have fitted the data in
question more closely than the second-degree parabola. But, es-
pecially if any attempt is to be made at extrapolation or the pro-
Jection of the trend into the future, the smaller the number of
constants involved in the formula the better. The more complex
curves are more likely to fit the data within the period covered,
but generally are harder to explain and less dependable in projec-
tion. Indeed, it would seem not unreasonable in the case at hand
that the exponential curve might really be the preferable line of
trend because more in keeping with the facts which formed the under—
lying reason for the movement in land values in the region. '

The fact that the trend in this case may .not unlikely have
been distorted more or less by temporary conditions, suggests that a
line of trend should be interpreted with caution. Ag new and more
recent data become available, the trend lines should be reconsidered
from time to time. Frequently it will happen that a trend line will
include more than one homogeneous period. 'lfter a change in direc-
tion of trend has occurred, it is likely to be some time before the
student can be sure either as to the fact of a permanent change in
direction or as to the amount of the change. Thus, as explained in
discussing the straight line, it may be necessary to divide a period
of time into shorter ones and to fit to each one a different trend,
and probably to use different formulae in the process.

181

Locating trends In their search for more rational methods of

b correlation - fitting trends, economists have turned to the
éEélKElfl devices of including a variable to measure

secular trend in multiple correlation anals '

ysis. Secular trend is usually taken as a
function of a series which increases one unit increment in magnitude
with the passage of each unit of time in the series. The residuals ‘
of the analysis are plotted around the regression line of trend deter—
mined in the analysis and a curve fitted to the residuals. Where a
large percentage of the variation can be accounted for in an analysis
of this sort, it is assumed that the effects of the other variables
are held constant by the method of multiple correlation, leaving only
the relationship between the variable and time. The deviations from
a curve fitted to the regression equation measuring the relationship
between time and the variable being studied are due to other factors
than those included in the study.*

Combination The objections to least squares can often be
methods* ' remedied by taking a three or fiveeyear mov—

ing average of the data and applying the method
of least squares to the results. In this way
the wide variations are eliminated and the ef-
fect of any extreme obserVation is spread through three or five years,
which tends to offset any undue influence it would have. Another ad—
Vantage of applying the moving average to data of this kind is that
its flexibility will reveal any tendency the data may have toward a
curvilinear trend, which might not be noticeable in studying a graph
of data with wide variations. Of course all of the limitations of
moving averages above described are pertinent here.

Another common combination of methods is fitting a curve to,
a moving average by free-hand method. In data having wide varia—
tions, one or two extreme observations may throw the moving average
out of line. Where sharp breaks or changes in direction occur, the
moving average is also thrown out of line; hence the necessity for
smoothing these departures from normal. Irregular length of periods
causes similar difficulties. With these limitations of a moving av-
erage in mind, and a thorough knowledge of the data being worked with,
it is often-possible to fit a smooth curve to the moving average more
accurately and with more consistency than by mathematical curves. A ‘
forecast made from such a curve will be on a safer basis than if upon
a mathematical curve alone.

Free—hand curves can be checked for accuracy in the same
manner as a straight line is fitted by the method of least squares.
The sum of the squares of the deviations cannot be made equal to the
smallest possible amount that is possible for any curve, because a

 

* IllustratiOns of this method of measuring trend are to be found in
U.S.D.A. Bulletin 1440 - Factors Affecting the Price of Hogs, G. C.
Haas and M. J;'B. Ezekiel and U.S.D.A. Technical Bulletin 50 - Fac-'
tors Affecting the Price of Cotton, 3.3. Smith, pp. 22-23 and 51—53..

** Contributed by Purves. \'

182

curve could be made flexible enough to pass through every point making
the sum of the squares equal to zero. However, with the general
shape of curve best adapted to the data in mind and the degree of
flexibility desired, it is possible to fit a curve of known shape

and flexibility so that the sum of the squares of the deviations will
be the smallest possible for that type of curve.

Seasonal The problem of seasonal variation. like that
variation* of secular trend, is to devise a method of

measuring the normal seasonal and then to

"eliminate" the effect of this type of fluc-

tuation from the series. Certain methods are
limited to measuring the ayerage seasonal influence, while others are
designed to obtain indexes which will reflect changes in the seasonal
factor over a period of time. Examples of both methods of measuring
the seasonal factor are included in the summaries given below.

The method of monthly means. This method of obtaining an
index of seasonal variation, although long in use, has received re—
newed attention.* The method has the advantages of brevity of comp
putation and simplicity of procedure. The procedure may be sums
marized as follows: Obtain a total for all the January items in the
series and corresponding totals for each of the remaining eleven
months. Sum the totals obtained and divide by 12 to obtain the av—
erage monthly total. Compute relatives for each month by dividing
each monthly total by the average monthly total. To eliminate the
seasonal factor, divide each item in the original series by the cor-
rGSponding monthly relative.

This method of course takes no account of changes in the
seasonal factor; Moreover, as a method of representing the average
seasonal, it has three important limitations. (1) The use of the
arithematic average allows an exceptionally high or low monthly item
to exertan undue influence on the determination of the typical value
for that month. Influences producing irregular fluctuations will
therefore be reflected in the seasonal. (2) If the cyclical move—
ment is large in comparison with the seasonal movement, the influence
of the former will tend to dominate and distort the seasonal index.
This tendency will be more pronounced in short series, although there
is no assurance that the cyclical will ayerage out over a longer
period. The seasonal index obtained is therefore subject to the in-
fluence of cyclical fluctuations in the series. (3) If the series has
a pronounced trend upward or downward. this also will affect the seas—
onal index. An upward trend of the series would make the January re-
lative too low and the December relative too hi ah. whereas a downward
trend would have the opposite effect.

- The considerations indicated in (1) and (2) may be regarded
as placing definite limitations on the use of this method. These
considerations suggest that the method is inapplicable to series in

 

*‘fificntributed by Clayton.
** Cf., e. g., Hart. William L., "The Method of Mbnthly Means for De-

' termination of a Seasonal Variation". Sept.. 1922, Vol. XVIII, :,
m 341-349 . c

183

which the monthly items are widely dispersed about the monthly mean
and also to series having pronounced cyclical fluctuations. As to
the considerations in (3), a further step may be added to the four
previously given to approximate the influence of secular trend. A
method*of doing this is as follows: Assuming the period covered by
the series is 1910—1919, subtract the item for January, 1910 from the
item for January, 1920 and divide the remainder obtained by twelve.
Add (or subtract) one-twelfth to the beruary total, two—twelfths to
the March total and so on for the remaining months.

Median-link—relative method (Persons). This method for
determining the average seasonal factor was devised by Professor
Warren M. Persons and**is widely used. The essential steps in the
computation of an index of seasonal variation by this method are
as follows:

(1) Calculate month-to—month link relatives for the entire
'seriee. ~

I (2) Arrange the link relatives obtained in a multiple fre-
quency table. - .

The closeness of the clustering of the items in each column gives an
indication of the precision with which the seasonal can be measured
by the use of an average. The displacement of groups between adja-
cent columns indicates the direction or rate of the seasonal change.

(3) Determine the median link relative for each month.
‘For short series especially, the three middle links should be aver-
aged to obtain the median value.

(4) Express the twelve median—link relatives as percent—
ages of a constant base by computing chain relatives with the January
link as a base.

(5) Correct the chain relatives for trend. If no trend is
present, then the product of the January link by the December chain
relative should be 100. When trend is present, the chaining of the
links has the effect of introducing a cumulative error, Compounded
over the twelve months at a rate equal to the rate of increase of the
secular trend. The rate may be determined by employing the compound
interest Imam”l specifically, as follows:

 

* Cf. King. W.I., "An Improved Method for Measuring the Seasonal
Factor",-Sept.§l924, Vol. XIX p. 301. Fbr the use of the least-
squares trend for the same purpose, See Davies, George 3..
Economic Statistics, p. 119. ' ‘

** Articles by Professor Persons describing the method will be
found in the Review 3: Economic Statistics, January, 1919,
pp.18—31, Journal of American Statistical Association, June,
1923, Vol.XVIII, pp.713-725, and in the Handbook g: Mathematical
Statistics, edited by H. L. Rietz, pp. 150—165. ‘

*** or. Mills, r. 0., Statistical Methods, p. 320.

184

January link x December chain . 100 (l—t'r)12.

‘ The chain relatives may then be corrected for trend by
dividing the February chain relative by (l‘f r), March by (1-+-r)2,
and so on.

,(6) The corrected relatives give an index of seasonal var—
iation on January as a base. To convert this to an index showing
the monthly seasonal as a percentage of the average for the year,
divide each monthly relative in (5) by their average.

The application of this method involves extensive calcula-
tions, and its use, in lieu of less laborious procedure.shou1d be
justified by compensating advantages with respect to the validity
of the results obtained. It has certain advantages over the method
of monthly means. (1) The median, since it is unaffected by large
or small chance variations, is less subject than the mean to the in-
fluence of irregular fluctuations. Except for series having special
characteristics,* the median Value is a more typical average than the
arithemetic mean. (2) As to the influence of the cyclical factor, it
is suggested that "there is an indirect tendency to minimize the in-
fluence of cycles because of the comparison, in the link-relative
operation, of each month with that immediately preceding. This fact
tends in general to free the link relative from cyclical disturbance.**

, (3) The correction introduced for secular trend has been
indicated above. '

‘Ratio-to-ordinatg method (Faulkner). This method for com-
puting an index of seasonal variation was devised by Dr. Helen D.
Feulkner.***

The essential steps in the ratio-to-ordinate method are as
follows:

1. Fit an appropriate line to represent the secular trend
of the original data.

2. Compute the ratios of-the original values to the ordin-
ates of the line of trend. . '

3. Form a multiple frequency table of the ratios.

 

* Cf. Hart, William L., loc. cit., pp.343-345; Persons, Warren M.,
loc. cit. p. 718; Crum, W. L., "The Use of the Median in Determ-
ining Seasonal Variation", March, 1922, Vol. XVIII, p.607-614,

** Faulkner, Helen D., lec. cit. p. 169.

*** loc. cit., pp.167-179. A similar method, in which the resi-
duals "obtained by subtracting the ordinates of secular trend
from the original data (are) expressed as relatives of the '
secular trend and then used in the calculation of seasonal
variation", devised by Dr. Lincoln W. Hall, appears on pp.156~
166 of the same issue.

185

4. Determine a typical-value for each month by taking the
arithmetic average of a group of (three to seven) ratios which in-
cludes the median and an equal number of ratios above and below the
median. '

' 5._ Adjust the twelve typical ratibs obtained in 4 by di-
viding through by their annual average to obtain a final seasonal
index having an annual average of 100 per cent.

Dr. Faulkner points out (loc.cit. p.174)_that this method
"makes adequate allowance for secular trend, for the trend is actually
eliminated from the original data before the computation of the seas-'
onal variation. Moreover, it provides against the influence of the
irregular deviations by using a group of middle items, in any one
month, from which the irregular extremes are omitted. The provision
against the effects of cyclical fluctuation consists chiefly in the
insistence upon a series covering a longrtime interval, but the
proper selection of the grouping used in getting the crude indexes
tends to minimize the disturbing effects of non—uniformity in the
cycleS". The method has the further merit of euoe’or computation:

The three methods Just discussed aim at obtaining an
average seasonal index applicable alike to all years covered by the
series. The fact that the seasonal factor is itself subject to
change has led to the develoPment cf methods designed to take ac-
count of the changing seasonal. Three methods of this type are
presented below.

The moving median method (King). In proposing this meth-
od, Dr. King's aim was "to derive an index which will show the trend
of the changes accruing in the seasonal factor from year to year. . .
It gets away completely from the crude assumption that seasonal fluc-
tuations are unchanging from year to year.*

The essential steps in this method are as follows:
(1) The original data are plotted.

(2) A preliminary cycle-curve is fitted to the plotted
points by freehand methods. To assist in locating this preliminary
cycle—curve, an approximate seasonal index obtained by the method of
monthly means is used.

(3) Figures for each month are read off from the pre-
liminary cycle-curve to obtain a tentative estimate of the cycle.

(4) The original monthly figures in (1) are then divided
by the corresponding monthly figures in (3) to obtain an approximate-
seasonal index for the various months throughout the period.

 

* King, Willford 1., 10¢. cit., pp. 304,305.

186

(5) To smooth out the irregularities, a nine—period moving
median is employed. The median January ratio for the first nine
years is written as the seasonal’figure for the fifth year and so on.
The process is repeated for the remaining eleven months.

(6) Monthly averages of the values in (5) are obtained for
each year. Each monthly value in (5) is expressed as a percentage of
the monthly average for the yen.r. The sum of these percentages should
be twelve. The curve of these percentages is smoothed to obtain the
required adjustment. The adjusted percentages provide a separate
normal season index for each year during the period.

(7) The itoms for each month may be eyeragod to obtain an
average seasonal index. but this stop is not of course contemplated
as an _object of the method.

(8) To obtain the final estimate of the cycle, each month-
ly item in (l) is divided by the seasonal index (6) for the corres-
ponding month of the same year.

Obviously, the results obtained by this method are con-
ditioned hy the Judgment and the technical «roficiency of the indiv-
idual in employing the freehand methods of fitting and smoothing
curves which the method requires. It is not probable that two come
patent persons applying this method to identical data would obtain
precisely the some results. It does not follow, however, that the
precision of results obtained by strictly mechanical methods would
necessarily yield a more, or even equally, accurate measure of the
seasonal factor. A method which nmasuros the seasonal factor with
approximate accuracy is certainly to be preferred to methods which.
yield mechanically_precisc results, but.which fail to represent ac-
curately.the seasonal fluctuation. A serious drawback of this '
method is that it necessarily fails to yield index numbers for the
last few years .of the period. The index is never "up-to-date".

The method, therefore, leans heaviest on approximations where the
consequences of error have greatest practical significance, namely,
for the period touching-the immediate present. - As’a method, it
helps least when help is most desired. It may be said, however,

that any method which accurately measures the seasonal for the

recent past will serve very well for current years, since the season-
al factor is relatively constant for short periods.

The method of adjusted monthly ratios (Gressens). This
method. like the method proposed by Dr. King, is "based upon the
consideration that yearly fluctuations take place in the seasonal".*
Mr. Gressens believes, however, that the greater reliance of the
present method on a definite mechanical procedure makes it superior
in this respect: to Dr. King' 3 method.

 

* Gressens, 0.; "On the Measurement of Seasonal Variation", June,
1925, Vol. XX, 1). 207. '

187

The following steps are used in the computation:

(1) Chart the original data. If the data exhibit no pro-
nounced trend, eXpreSs each monthly item as a percentage of the aver—
age monthly value for the particular year. If, however, the data
exhibit a pronounced trend, these ratios.do not sufficiently allow
for the monthly trend, and must.be corrected. An alternative pro-
cedure is'therefore recommended when an inspection of the chart of
the original items reveals a.pronounced trend in the data. This con~
sists in expressing each monthly item of the original series as a
percentage of the corresponding monthly-ordinate of the secular trend
of the entire series. The ratios obtained are free of the annual
monthly increment_due to trend, but are still subject to the influence
of irregular and cyclical fluctuations.

(2) The monthly ratios are therefore smoothed by using a
moving average —— mean or median --'of an appropriate period. In the
illustrative example a five-period moving median is empIOyed.

(3) Any irregular items are then moderated with respect to
adjacent items for the particular month.

(4) The smoothed ratios are extended forward and backward
to obtain terminal values for the series and are adjusted so that the'
sum of the ratios for each year is equal to 1200. '

Other methods. The methods summarized will sufficiently
illustrate the procedure employed.in handling the problem of seasonal
variation. For the most part, these methods are directed to the prob~
lem of characterizing the seasonal movement of a series with a view to
the ultimate isolation of the cycliCal fluctuation. A method directed
more specifically to the study of the changes in the seasonal factor
over a period of time-with a view to the development of seasonal stand-
ards has been devised by Jrofessor Crum*. When applied to series in
which Professor Crum's modified monthly mean device may appropriately
be employed, the method admits of a direct and simple application to
isolate the cycles (Ioc. cit., p. 60). The process does not require
separate calculation and elimination of the trend and saasona1.'fl
But for series requiring the use of the modified link relative method.
the elimination of the seasonal factor from the series would'prove
rather laborious, even though the "centering" process proposed by
Professor Crum for the determination of the seasonal for a given,
date is employed.

 

* Crum W. L., "Progresaive Variation in Seasonality", March, 1925,
Vol. XX, pp.48-64. ‘

** Cf. Brumbaugh, Martin Allen, Direct Method of Determining Cyclical
Fluctuations of Economic Data. Dr. Brumbaugh proposes a method,
based on certain special assumptions, for isolating the cycle in
which, Also no attempt is "made to segrate and measure either'ths
trend or the seasonal variation". (loo. p. 15) K

188

CYclical There remains to consider the problem of
Movements* separation of cyclical episodic and chance

fluctuations. As a matter of practical

observation, it is generally impossible to

separate these frdm each other. All that
is usually done-is to find the residuals after secular and seasonal
movements are removed and then to examine these to discover evidences
0f CYOIGS. This process is illustrated by Table IV. Here the re-
siduals are expressed as percentage deviations from the corrected
secular trends. Vivid swings back and forth across the trend line
appear at irregular intervals. An examination of the data on the
size of the corn crop in the United States'shows that the outstanding
reason for these variations is in the fluctuations in the production
of corn from dry or wet years.‘ A second influence is found in the
number of hogs in the country, which determines to a large degree the
demand for corn. Thus in 1905 and 1906, there'were large corn crops
and the resultsrare seen in the variations of the price below the
trends in the periods Just following. In 1908 there was a moderate
sized corn crop, but a large hog crop to consume it and consequently
high prices. In 1909, 1910 and 1912 there were.large corn crops
and conseQuently declining prices. In 1911 there was a small corn
crop and a large hog crop. and consequently a high price.

It might be said that the variations just pointed out are
entirely attributable to episodic movements. Or it might be said
that the irregular wave—like motion which runs through the series is
cyclical fluctuation, while the more abrupt and short-lived varia-
tions are episodic in type. Assuming that the latter is correct,
still it will be impossible to separate the two types of variation
or to measure either one without making some very arbitrary'as-
sumptions. Making these assumptions, the short—time fluctuations
of two or three months in length may be smoothed out and the cyclical
and episodic deviations made to stand out-more clearly by means of a
moving average. ~-Usually a.short—time average of 3,5 or 7 months is
preferable as not smoothing out too many of the shorter episodes.

The resulting wave-like curve shows the main movements of the prices
which may be studied, each separately. in the light of the record of
the crop yield and the history of related enterprises. Since these
fluctuations seem mostly attributable to the fawcrable or unfavorable
seasons, it may be possible in this case to arrive at some general
statement of the relationship between the season and the fluctuations
in corn prices. But in episodic movements of most other types, this
is not likely to hold true and each episode may be caused by differ-.
ent influences and therefore call for a separate study.

 

* Contributed mostly by Hopkins.

189

Table VIII. Cyclical Variations in the Price of Corn,

 

 

' "1904-1914.
1 ' 2 3 - 4 5
.Seculaf Sec. Trend Price-i~ % Deviation
Price Trend Cor. for Cor.Sec. from Trend
»Season Trend

1904 Jan. .451 . 43': ' .438 .1.03 «05.
12‘. .501 .499 .455 1.10 +.1O
Mar. .52? ’ .430 .480 1.10 1.10
Apr. .527 .492 . . .472 1.12 r.12
Mby ’ .486 .494 .494 -.98 ~.02
June .488 .495 .490 1.00 .0'
July . 486 . 497 ' . 512 . 95 -. 05
Aug. .535 .498 .538 .99 —.01
Sent. .528 .500 .550 - .95 -.04
Oct. .536 .502 .522 1.03 +.O3
Nov. _ .541 .503 .518 1.04 - +.O4
Dec. .452 .505 .485 ~ .95 - —.O5

1905 Jan. .426 .507 .456 .93 -.07
Feb. .441 .508 .472 i .93 —.07
Mar. .470 .510 .500 .94 —.04
Apr. '.478 ’.512 - .492 . .97 -.03
May .562 ' .513 .513 1.10 +.10
June .542 .515 .510 1.06 ..06
July .534 .516 .531 ' . 1.06 +.06
Aug. . 50 .5;8 .559- .98 —.02
Sept. .528 .52 .572 .92 —.08
Oct. .522 .521 .542 ~ .95 ' —.04
Nov. .435 .523 .539 .90 3.10
DOC. .451 .585 .504 .92- —.08

1906 Jan. .430 528 .473 .89 —.11

 

Note: The foliowing years are similar to the two shown above. Since
the purpose of this table is illustrative, only the first two
years are shown. '

Periodicitz* Recent students of periodicity in economic
series seem to think that true periodicity
with the possible exceptional seasonal var—
iations is not to be found in economic data.**

Professor Crum writes "In economic science, barring the case of season-

al variation which is sometimes for short intervals very closely por—

 

* Contributed mostly by Killough and Wertz. - .
** Day, E.E., Statistical Analysis,(1925)0hap. XIX.‘p.306; Cnmn,W.L.,
and Patton, A.C., Economic Statistics,~p. 384.7

190.

iodic, true periodicity is unknown.* It is also probably quite ques—
tionable as to whether an exception need be made of seasonal variation, f
for it is very doubtful whether economic values repeat themselves in
exactly the same months, or weeks and certainly very seldom on the

same days. The price of eggs is usually higher in December than in
January. but this not not true invariably. It certainly would be
futile to adhere rigidly to a definition of periodicity in the treat—
ment of economic data. Through absolute periodicity is not to be
found in economic data with the possible exception of seasonal variaw
tion, it does not follow that it is not important to study more-or—
less periodic movements. Much has been gained by the study of correl—
ation between the series, yet very seldom, if ever, is perfect correl-
ation found. It is important, for example, for the former to know in
what month or months the price of wheat is usually highest and the
price of commercial feeds lowest,ovon though this average does nor re-
present the true situation for every year.

If it is assumed that the fluctuations are of a cyclical
nature, it will probably be desired to find out as much as possible
about the lengmh of the cycles and their amplitude. It is very sel—
dom that a series of economic cycles is encountered with anything like
a uniform length. Regular periodicity is therefore hardly to_be hoped
for, and it will be necessary to ascertain the length of each cycle
separately. There is no certain and fixed method for such measurement,
largely becauso we are not sure as to Just what constitutes-a cycle and
where cyclical variations leave off and others begin. The-degree of
regularity of'time intervals in a given series of data may be determ-
ined by measuring the intervals between successive crests or between
successive troughs of the plotted data and by tabulating the results '
in'a frequency table. Better than either of these, the length may be
counted in months or years from the point at which the cyclical curve
cresses the trend line until it crosses it again in the same direc-
tion. This method applied to the data of Table IV after they are
embothed by a 7-month moving average would give three periods of
57.21 and 33 months respectively.

Such variations in the length of cycles are not unusual.
They illustrate very well the difficulty in the way of discovering
any typical or normal length of cycle. Attempts have been made to
construct a typical cycle for various series, but in most series
it has been without much success. In the case of the hog—price
cycle, which is regarded as one of the most regular in agriculture,
the average length of six cycles from 1880 to 1913 was about five
and a quarter years, but the individual cycles varied from 113 to
37 months. No cycle in the whole period was within six months
of the average length. The idea of regular periodicity is there-
fore quite untenable even for this series.

 

* Crum, W. L., "Cycles of Rates of Commercial Paper", "Review of.
Economic Statistics", Preliminary Volume V.. p. 17.

,_.1:' ., ;

The study‘of the cyclical and-episodictvmaeions will there-
fore need to go deeper than to postulate a regular oscillation equal to
the average cyclical duration. Any worth while analysis will involve
a careful study of events which accompany or precede theSe variations
in related series, or which, perhaps, are not expressed quantitatively
in any series. -

There will also be a wide difference in the amplitude of ‘
different cycles as well as in their length. There will be as great
a difference in the amplitude of the variations in different series;
For instance, as compared to the variation of 25 to 30 per cent above
and below the line of trend in the corn price series from 1904 to
1914. the variation in the index of all commodity prices in the same
period was only 8 or 9 per cent. 1

The simplest method for comparing‘the cyclical fluctuations
of different series of data is to plot them for visual comparison of
regularity and of duration and amplitude of fluctuations, and for
visual estimation of timing or lags. More precise and complicated
methods are as follows: '

1. For comparing amplitudes of cyclical_variations in
different series of data, standard deviations may be computed of the
relative deviations of each series from lines of secular trend.*

2. Another method for comparing amplitudes of fluctuations
of the several cycles of a series of data has been used by F.C. Mills.**
Mills expresses the rise of each.cyc1e as a percentage of the ensuing
high, the fall as a percentage of the preceding high, and calls the
average of the two results an index of cyclical variability. An ad—
vantage of this method is the fact that it obivates the necessity of
eliminating secular trend. '

3. Comparative.t3ming or lags of different series may be
expressed as deviations from selected reference dates either in the
beginning of revival or in the beginning of recession.*** Another
method sometimes employed to measure lag is to compute several co—
efficients of correlation between two series, pairing the items
with different lags. The lag which gives the highest coefficient
is taken as the most representative.**** '

. 4. The average_conformity of variations in time series
in direction, duration, amplitude and timing may also be measured

 

* See Mitchell. W. C. "Business Cycles: The Problem and its

. gSetting", 1927, pp. 278—278.' ,

** Mills, F. 0., "The Behavior of Prices", Publication No. 11
of the National Bureau of Economic Research, p. 84.

*** Mills, F.C., "The Behavior of Prices", Publication No. 11

' of the National Bureau of Economic Research, p. 92.

**** Mitgggll, WLC., "Business Cycles: The Problem and its—Setting“,_ ,

Po 0 ' I ”‘1‘

192.

by correlation. The use of this statistical method by one who is
not familiar with its limitations, especially as applied to time
series, may lead to erroneous results. If used with judgment, hOw-
ever, correlation is a very effective method for comparing the aver-
age conformity between the deviations of- different series of economic
‘data. In the correlation of time series for the purpose of compar-
iné‘CYC11Cal variations, the data to be correlated are commonly re-
duced to the form of variations from secular trend or of first diff-.
erences. ' '

If two or more unlike series (crop yields, for example) are
found to have similar cycles emanating from common causes it may be
desirable to make a synthetic curve by combining the several series
into one composite curve. ' For this purpose it may be necessary to
eliminate secular trends of each Series by expressing the residual
cycles as deviations from trend. Theso deviations may then be ex-
pressed in terms of a common unit (standard deviation) before the
series are added.' This rather complicated process tends to equalize
the weight given to each series of data entering into the synthetic
curve.~- , ‘

Numerous studies of the business cycle have centered on the
problem of demonstrating periodicity. Professor H. L. Moore, in a
series of. articles appearing in the Quarterly Journal of Economics."I
attempts to establish an e ight-year period for the business cycle
based upon the theory of periddic generating economic cycles whose
.ultimate cuases are to be traced to meterological effects due to the
.periodic motion of the planet Venus. What first challenges the
attention in the results obtained by Professor Moore and others is
the wide variation in the conclusions reached respecting the typical
duration of the cycle. ** The conclusions reached by Professor Moors
Thave been challenged*** on the score of the periodogram analysis
which is employed. It is characteristic of most economic series
that they are relatively short and that their fluctutations vary in
amplitude, timing, and shape. The poriodogram analysis is quite
unadapted to the study of such series.**** The fundamental limita~
tion of the .periodogram analysis appears to be that its use implies
the assumption that periodicity exists. ***** A further limitation
upon the use of periodogram analysis (which, as noted. is applicable

 

* Generating Cycles of Products and Prices", February, 1921. Vol.35,
pp. 215-239; "Generating cycles Reflected in a Century of Prices",
August, 1921. Vol. 35. PP.503-528; "The Origin of the Eight~Year
Generating Cycle". Nov., 1921.Vol.36,pp.1-29.

** Cf. Thorpe and Mitchell, Business 43mm, pp. 38—45.

**‘ "All data could be explained as well by a theory of cycles which
merely assumed that in general the length of cycles is close to
that of their ayerage length, showing no true periodicity". In-
graham. Mark H., "0n Professor H. L.Moore's Mathematical Analysis

of the Business Cycle", June 1923, Vol. XVIII p. 765.
**** Cf. Crum and Patton, 10c. cit., pp. 381-383: Also Professor Crum's
article in the Handbook of Mathematical Statistids ,especially,
. pp.l74—176.
***** Cf. Professor Wesley C.Mitche11's discussion in Business Cycles_,
pp.259-261, and the reference there cited.

193

only to lgng series) to establish periodicity of the business cycle
is implied in the hypothesis advanced by Dr. Mills "that the dura~
tion of business cycles in a given country is a function of the
stage of industrial development which that country has attained.*
The obvious inference from this hypothesis is that periodicity is,
at best, characteristic of the business cycle only for relatively
short periods, that is only for those years in a country's history
which represent a relatively homogeneous, phase of industrial de-
veIOpment.

The pertinence of this consideration when applied to the
fluctuation of individual time series is obvious. The character
of the cyclical fluctuations in price series for such products as
hogs, beef-cattle, and horses is obviously conditioned by changes
of a structural sort in the economic order. The hog cycle, for
example, was longer in 1870 than at present. and pronounced mod—
ification of the horse cycle is new apparently under way. Observed
fluctuations in a series are to be regarded as the net result of
numerous interacting and mutually conditioning forces. But the
interaction of these forces, and hence their net effect, is con-
ditioned by the structural characteristics of the economic order
at a given time.

£95** One of the pitfalls of research with time

- series data is the matter of lag. First of
all, its presence must be actually demon-
, stratod. Second is the task of measuring
the amount of it.

Lag is defined in the dictionary as the "falling behind or
retardation of one phenomenon with respect to another to which it is
closely related." While this definition comes from the use of the
term in the physical sciences, it is equally applicable to economic
terminology. Lag as thus defined may exist between phenomenon
whose relationship is eith r qualitative or quantitative. The mere
presence or absence of one phenomenon may be dependent upon the
presence or absence of another at some preceding tine, or the quan-
titative measurement of one may be dependent on the giantitative
measurement of the other at some preceding time. The economist
first of all is concerned with discovering whether the presence of
one phenomenon is coincident with or lagging behind the presence
or absence of another. In the more recent studies, however, em-
phasis is being given to the measuring of quantitative relationships.
and the following discussion is made primarily with those studies in
mind. '

 

* Mills, Frederick 0., "An Hypothesis Concerning the Duration of
Business Cycles", Dee. 1926, Vol. XXI, p. 447.
** Contributed mostly by Mr. Felt.

M
‘v

194

1-'

/~/

To say that one phenomenon lags behind another assumes a

causal relatidnship or path of influence between the phenomenon.
These pathh of influence vary from the simple d.irect cmuw -..nd-
effect relationships to complexities beyond exclanation. Thus we
find thot Variations in the weather from year to year produce varia—
tion: in the _yield of potatoes and. finn.lly variations in potato
prices. The nath of influence e"1s*11’ Jet een ootato.yie1ds end .
price is direct and reletively simile. Far hermore. the relation-‘
Shiv) aresents no difficult -1roblem of lag determination. A fluctua-
tion in potato prices, however, thru its effects on profitableness of
notato iroduction may affect subsequent potn,to production, maybe for‘
mtly one year, may -e for several yesrs _oossibly in one the first ”' '
ye.ar .1nd in the onposite the next. The n.th of influence in this
6 .se is less simple and the problem of lag determination is ves9 come
let. Again, given fluctuations in the weather may produce changes
in the prices of two different croos, on one of them more delayed and
,rolonged effects, amounting to n 1::. 9 he price of influence be-
treen the tn 0 rel .ted orices is OJ"lOlSly indirect in this case. In
this case, the nature of the caussl relationship is known: but often
the common generating factor may be unknown, or else it may appear
that the path influence is direct. ‘

It is not safe to conclude that the changes which appear in
one phenomenon are direct effects oi the changes in another, even though
the former may be predicted with re.1sonable accuracy from preceding
'Ch-tnses in the latter. This does not men n, however. that these relan
tio=.1shi;s are not Worthy of study. But while it is sometimes necessary

1;to deal with ligged rele tionships betv.een Series for which the oaths of
influence are not known, these instances “re likely to be dangerous for
forecasting purposes. '

Leg is inevitable between econoznic phenomenon. particularly
in case the path of influence :ivin he relationship is one of direct
c muse 1nd effect. Fen 1f "my econo.ic adjustments c1n be made instan—
taneously, is time is required for the social orginism to act; One
~1ine oi ‘uusiness- nay become suddenlv profitable; yet before men and.

c.Tjital are_ "ttrected to its orofitibleness trust become. known and
decisions must he made. These huno.n processes all require time.'
lirs'i'r there are nhvsicil 1nd biolo: ical factors limiting the
Spflfifl 0f .adjnstmentsc s?1ctories-1ey need to be constructed or live-
lfiufik raised before the final product will as forthcoming. _ 1
I ' I
lbile some lag is inevitable between most 'eoounnso ghcnwmnnnu'
the time lumen-271:3 uncertain. Any thing which will speed up abuse;
'mnts will tend to shorten this lat: period and. anything Emmiimg men
adjustments lengthen the lag period. Thus. in attempts to measure lag,
vit is important to keep the causes of the lag in mind and their effect

on the_rate of readjustment.

In some cases the problem of lag involves determining the
time of the reaponse to a single event or a succession of discrete
'events. The effect of tariff chenies on the prices in an importing

195

country may serve as *n illustration of discrete relationships. The‘
crop and price rel tionshio also illust'rntes it An example of the
continuous type of 3.01nti0n5hip is the dairy-product and price relaé
tionship. The rainfall-yield relationship is Continuous during the
growing season ReletionShip during: such a period may he in the .
same or opposite directions - Thus, it may well be that heavy rain~
fall is favorable_ to crop condition durin: port of the period and un-
favorable at othe.r tiaes vuthin the -period.

. It is necessary to reco nizc an important type of the con—

tinous type of reletionshio which may he desisnated as cumulative.
Continuous relationships frequentl.y aieretc in the same direction .
and in such a manner that the cwnuletive effect of one series fore—
shadows or causes the changes in another. To cite a simple illustra-
. tion. the daily balances of ones bank account is simply the cumulative
of the net deposits during all precedin; time. Other relationships. ‘
not so exact. tend to operate in like manner. For instance, Xarsten
finds that the price-leVel of securities shows the cumulative effects
of the flow of money (into the market out of‘ business, or out of the
market into business).

Lag between two related series may exist fer any one or more
of the numerous types of variations found in the. time series, the .short-
time erratic variations the seas onel v.1riations, the waveolike cyclical
variations the episodic varia.tions due to wars. panics. etc.. and the
long-time variations or trends. Connarisons of daily changes in the
price of a commodity in different markets or p.t different steos in the_

. market channel very commonly show tend.encies for lag.‘ Thus the price
of wheat at country elevators is likely to lag one dry or more behind
the terminal market price. Much of course do acnds on the method and
frequency of obtaining market informa.tion and on the nature of local

competition.

There appears to be some misunderst~z.dinr as to the use of
the term lag where two role] ed series. are being comuared hy alotting.
or by correlating. This misunderstandin: results from the fact that
,the term is used. first to describe the sequence of events between ’
related series and second to descri“e the process of adjustinr- -the
‘series in comparisons by plotting or corrrelntinr. To illustrate,

let it be assumed that fluctuatidns in series A cans e fluctultions
'in series B, the latter following the former 1W some definite period.
say three months. In this instance probably 11 would a~ree thgt
series B lagged behind Series K. But in cmse the t .0 series er
plotted or correlated so as to allow'for this leg, which.series is
"lagged" in the.plotting or correlatino:nrocesai The most logical
answer to this queStion, and the one‘which'seems to agree with the"
terminology of most rtiters, is that the A series is lagged. It

is this series which is moved to a later -p.=riod of time- so as to '

compare it with theB 1 Beeries.‘ Since 3, as to sequence of events,
naturally lags behind A by three months, the letter is legged an
equal amount to bring the two series into_juxtaposition. Since

the 3 series is the result or the dependent variable, it is the factor
which is to he forecasted and. must retain its original position as
related to time.. . .

The importance of qualit. tive- maiysis 1n prohdems of lag
cannot be overstated Given a three-yer r cycle in which two series
. are opoosite each other, so far as the mathematics of it is concerned
it can be equally well said that A affects B with a lag of eighteen
months, or that B affects 1 with the s me 1a;:. Another excellent
example is Figure 4. riven earlier. Here atricultural prosperity
is made to appear to 19.;( behind business prosperity by the simple ,
procedure of assumin; a lap of 3 to 4 years Suppose we were tckmake
a SimPler and more rational assumption that for the most part agri~
culture and business ought to pr05per to; ether That would mean.
drawing -perpendicular rather than slanting- lines. It would then
apeear that business and aariculture have been good and had together
almost two-thirds of the time. A Careful study of the occasions
’when they have not been, as in 1905-8, 1914, and 1922-4. would then
nodou‘ot reveal some significant further interrelationships. In the
same two series, the causalrelationship mi:ht be made to run in the
1 opposite direction by the simple device oftslanting hhe lines the
opposite direction. In fact, a conclusion1closely related to this
was the first one reached in this study. (See the J. F. E.. July,
.1925, p. 361—71 and p.479-85 in October number ) Coupled with the
foregoing are the inaccuracies already pointed out growing out of
the use of moving averages.

. Following are some of the methods of measuring-lag that are
in use: a ‘
'l. Graphic p See Day; p. 324-7.

' 2. Simple correlation of the two series to seenhich lag gives
the highest coeffici;nt. See Rictz, ppgleo- 65; Crum and
,Patten pp. 364-370; 2dills, pp. 420.430,

3. multiple correla.tion of tvzo series and major other factors '
affecting the larc;inr series, See Ennherg's "Industrial
Prosperity and the Farmer" for an example of this; also re-
view of same in June 1928 Journal Am. Stat. Assn.

45 More mathematicp.l methods are the quadrature method used by-
Kersten. and thefactor system used by Zinn.

, It should not be forgotten at any state of such study that
the mathematical 01 mechanical processes which have been described d;
not in themselves explain any of the trends or lags which they bring
to light. They are simply mechanical methods which facilitate anal-g
ysis and may be applied in discovering or describing the behavior of
. the series. In studying each series, the apoarent reasons for the
- lags in question must he kept in mind throughout the process. In each
case, somewhat different interpretation will .30 applied to the lags
discovered depending on the nature of the series and the relationships
whiCh seem to exist between them and related facts,

KL

 

 

 

 

flffifih r—‘jfi Ht“? W‘yTnnflr- Hqflmfi
Bulimia“ all»... Lilli-aw lg}, Lady-ll ."l

Research Method and Procedure

in Agricultural Economics

A Publicationpf the Advisory Committee
on
Economics and Social Research
in
Agriculture
of

The Social Science Research Council

VOLUME TWO

AUGUST 1928

 

 

 

The material in this volume Was prepared under the
direction of the AdviSOry Comitteeon Economic and Social
. > Research in Agriculture of the Social Science Research
._ ‘~' Council. The constituent orgmizatione of the Social
’ ~~ \ Science Research Council are as follows:

American Economic Association
American Political Science Aesociation
American Sociological Society
American Statistical Association
American Psychological Association
America; Anthropological Association
AmeriCan Historical Aesociaticn

The members of the Advisory Committee on Social and
Economic Research in Agriculture are:

H. C. Taylor, Northwestern University, Chairman
John D. Black, Harvard University '
Joseph S. Davis, Food Research Institute

C. J. Galpin, Bureau of Agricultural Economics
L. 0. Gray, Bureau of Agricultural Economics

. E. G. Nourse, Institute of Economics

.1 r G. F. warren, Cornell University

TABLE OF CONTENTS
Vol. Two

Part II. Methods (Continued)

g. Analysis of Data (continued)
(3) Association Analysis, p. 197
(4) Correlation analysis, p. 219
Elementary analysis-cross section, p.219
Elementary analysis-time series, p. 232
Variables to be included, p. 232
Statement of variables, p. 233
' Choice of method, p. 23’s
Significance of results as description of
data, p. 241
Effect of/errors in data, p. 249
correlation as description of geographic data,p.258
Multiple correlation mechanics, p. 258 '
h. Inference fromrfESults, p'271.
(1) Statistical'inference, p. 271
(2) Inductive inference, p. 275
(3) Inference‘from correlation in time series, p. 285
(4) Predicting from correlatiOn results, p 288

,B. The Method of Analogy, p. 298

C, The Case Method, p. 300

D. The Informal Statistical Method, p. 309
E. The Experimental Mbthod, p. 314

Part III The Research Approach, p. 323
A. Cross-section — non geographical. p. 324.
B. Cross-section - Geographical, p. 329
C. Time sequence - Historical, p. 341

Part IV The Presentation and Utilization of Results, p. 352
A. The use of tables, p. 352
B. Graphic presentation, p. 360
0. Preparing the written report, p. 383
D. Utilizing the product of research, p. 387
(over)

Part V Organization of Agricultural Economics Research, p.408
Viewpoints in economic research, p. 403
The localization of research problems, p. 405
The localization of_researdh facilities, p. 406-
Research policy ana program, p. 408
The classification'of'roseardh problems, p. 414
Research outlines by fields, p. 415
Relationships between research in economics and in other subjects,4l7
Relation of research to other functions,‘p. 423
Administrative arrangements relative to other functions, p. 426
' (a) Research in related departments, p. 426
(b) Research dnd related functions, p. 428
(c) Research in one department, p. 434

Accrediting results, p. 435 ' . ' '
Relationships to resaarch in other institutions, p. 436
Relations with nonrcoordinate.agenciesg_p. 487
Cooperative procedure, p. 438 - ‘
OVerhead organization as affecting research is
in agricultural economics: P3 452
(1) Scheme of organization, p. 452
(2) Allocation_of funds to economic research, p. 456

Adnflnistration, p. 462

(1) Selection of research staff, p. 462
(2) Provision for training of graduate
- .studenta, p. 465 i
(3) Sabbatical leave, p. 465
(4) worm of service, p.,466
(5) Division of time between teaching and
research, 9. 468

R E S E A R C H M E T H 049 A N D P R 0 G E D U R E

IN AGRICUQTURA; ECONOMICS

V 0 L U‘M E VT W O

(3) AssocmxonimmsIStV

The analysis of data by methods-of aseociation offers a log-
ical approach to an explanation of correlation analysis. Most workers
in the field of agricultural economic research are familiar with meth-
ods of correlation; but many are unaware-that by means of the cross
classification of data and an analysis of the percentage of cases
which occur in various combinations, it is'possible to discover and
evaluate aesociations of either variables or attributes without resort—
ing to correlation methods. It will be the aim of this section to
present the fundamental principles of association analysis, and to
follow, with illustrations which show the analogy between the results
of association and-those of correlation. v'

The fundamental considerations of aseociation analysis may
be'emplained better from a simple illustration of hypothetical data
than from actual data. For this purpose, let us assume that records
are aVailablerfor-twenty farms in a given district which show whether
each is an owner or tenant, and also whether each has or has not an
automobile. If these data were tabulated according to both character—
istics, a cross classifiCation table such as the following may also be
asSumed: '
‘ Automobile Ownership

 

Tenure Number who do Number who :

 

 

of farmers : not have autos : have autos : Total
: Ab : AB : A
Owner :: 9 E 6 : l5
: ab : a3 : a
Egnant : 3 : 2 : 5
: b : B : N
Total : 12 g 8 : 20

 

In these illustrations, capital letters, A, B, C, will be
used to denote positive characteristics. The opposite case, where
the characteristic is negative or lacking will be designated by the
corresponding small letters a, b, c. In the hypothetiCal data cited,
owners will be given the Symbol A, tenants (non—owners), the symbol a.
Those who have automobiles will be designated by B, those who do not
by b. Two of these symbols used together will designate the combina-
tion of characteristics which is accomplished by cross—classification;
13, the number of farmers who are owners and have automobiles; Ab, the
number of owners who do not have automobiles; a3, those who are tenants
who have automobiles, and ab, those who are tenants who do not have
automobiles. Since the total number of owners (A) WOuld be made up of
the owners who have automobiles plus those owners who do not have them,

 

' borothea Kittredgp, University of Minnesota.
(197)

198

AB plus Ab would be equal to A. Likewise, theinumbers represented by
ab plus a3 would equal a; AB-+ aB - B; Ab 1 ab = b, These varimus_
symbols have been entered in the upper left—hand corner of each com—
partment of the cross—classification table, to assist in the analysis.
Any number of cases which can be designated by a single letter is known
as a first-order frequency and those designated by a combination of two
letters as a second-order‘frequency. The test for association in any
cross-classification table like the above is made by a comparison of
percentages or proportions of the cases found in different combinations.
Formulas are unnecessary if one thinks readily in terms of percentages;
Otherwise, simple formulas for various percentages, expreSSed in the
foregoing symbols, assist materially in carrying through an association
analysis. ‘

let us assume that our interest in the hypothetical case of
the twenty farmers, is to ascertain whether owners are more likely to
have automobiles than are tenants. Of the total of fifteen owners,
six or forty per cent had autos; of the total of five tenants, two or
forty per cent had autos. This is equivalent to saying that the number
of owners having autos was to the total number of owners, as the number
of tenants haying autos was to the total number of tenants. 'The pro-
pertion in terms of symbols would be ABSA: :aB:a. Of the total of
fifteen owners, nine or sixty per cent, did not have autos; of the
total of five tenants, three or sixty per cent did not have autos. The
number of owners not having cars was to the total number of owners, as
the number of tenants not having cars was to the total number of ten—
ants (Ab:A::ab:a). In the two tenure groups, there is exact agreement
in the percentages (forty per cent had autos, sixty per cent did not).
These statements are an indication of total lack of association between
the tenure of these farmers and automobile ownership,—— a condition
which Yule has called "complete independence". Inequality in the per—
centages according to auto ownership, between tenure groups, would be
evidence of association.

Since the degree of association present in all cross—classi—
fications of this general character must be Judged by the manner in
which they differ from the case of complete independence, the criteria
of complete independence must be developed at the outset. The first
of these criteria is that which has already been observed from the
proportions, that the percentage of automobiles among owners of farms
is equal to the percentage of automobiles among tenants, in which case,
Q._a_.§ <§___.g; 4o%=4o%>
A a ( 15 5

In the same way, the percentage of farmers without autos among the
group of embers is equal to the percentage without autos among the
group of tenants; .

$3.933; aw: 60%=60%
A a 15 5

~This criterion of independence mdght have been illustrated
equally well if the problem had been considered from”another-angle.

199

If the interest had been in “.scertnining fr0m the standpoint of quto-
mobile onnershi3 the 3ro3ortions which vere owners or tenants the oasic
'groups would have Been B and b and the following 3ercenteges would have
been examined:

13:42; '6=_9__;75%-_= 75%
B b 812

ea - g3; g3 §_; 25% =  25%
B b _ 8 12‘

. These qualities show, respectively, that the percentage of
farm owners among those Who have autos is equal to that of the owners
among those who do not have autos, and the-t the percentage of tenants
among those the have autos is equal to that of the tenants among those
who do not have autos. A situation of couplets independence is found
in an analysis made from this viewooint as well as from the one first
3ro;osed Alt11ough the analysis for 1ssocietion may be made on either
the- basis of rows or columns, usually the st tenent of the problem in—
dicates which is the 10 T1031 fonn for a given . ase. Except for purpose
of illustration, only the percentages on the Ac and a universes would
have been cor puted in this case. '

The first criterion of conjlote independence between the two

main characteristics of a cross-c-1.9ssificetion is, therefore. estabe _ *
~lished if the 'following percentefes ire eQual: -
12:11 seas:
A n ' _ B b
or '
11.21 11.11
A ‘ a ' B b

.The second criterion of comnlete independence may take a
variety of ferns, Thus far the -percenteges have been computed on
’the b1sis of the A. a, B, or b uni ivorses, but the s—econd criterion
involves also the greater univ rse represented by the total of N,

-or 20 in the hy3othetical illustration. If no association existed
- between ovmership of farm and automobile ownership, then we Ihould;
expect the number of owners with sutos to bear the Same relationship
to the total number of ox.ers .13 the total number having autos does
to the total number of fa ere, or ex3res ed in symbols AB: A::B: N.
In the illustration this "ould be 6:15::8:20, which indicates again

th1t no association is present This proportion is equally. true
when expressed- in any of the following forms:: .-
1939.3: ' L331) 533.222.; g-Q;
.1- N 13' N)’ N' 11 NN

The ebove proportion in whichever of the forms it is ex-
pressed will form an equality only' in the event of comjlete indepen-
dence in the data. Ineqaulities exhibited in the proportions are

200

indications of associetion. Conclusions from the first'two of the

four statements of the fireportion (1.6. g; _ 3 and fig _ %) are not
' ' ' "A N B ‘

as reliable when applied to data as are percentages of the tyre

£92.83. OTABJAb

A a B b

That the inequality between a statement of the form in-
volving é§_end A2 would furnish a better criterion of association
. B b ' .
'than the more general form involving AB and A is true'especially in
3 F
a case where the majority of the oojects or individuals in the un-
iverse are B's; i. e., if g a; TOtChGS unity, A; will necessarily

. N 3 B
iigroach A even though the difference between Ag'and Ag is consid—
N A - B b '

able."I

The third statement of the proportion, A3‘u‘A.E. is exceed—
. ‘ N
ingly useful, and is frequently used to determine the nature of the
aesociatiOns. A3 equals An B. in the case of complete independence,
N .
as 6 . 15 x B in the above illustration. If AB were greater than.the
20 ‘
fraction-gézgfi-thore would be evidence of positive association between
A and B. nr if AB were less than the fr1ction, negative association.
The lest of these four statements, an. A T. gives the rule that if -
'.\T NN ,
A and B are independent, the aroportion of AB's in the universe is
equal to the proportion of 's multiwlied By the 3roportion bf B's.

The chief advantage of the stn tements given under the first
criterion over those listed under the first criterion is that they _
give expressions for a frequency of second order in terms of the freo
quencies of the first order and the whole number of observations,‘
while the statements of the first criterion do not.

The criterion of independence ney'assume a third form, which
involves only second-order frequencies, and enables_one to recognize
almost at a glance whether or‘not the two attributes are independent,
and if not, whether the association is jositive or negative. ,

we have developed the stetement under the second criterion
that A3 a ‘ B' in the event of complete indefiendence. A corresponding
-fi_ ' ' .
Statement, for the ah's could he'develo?ed along exactly the Same lines,
namely. ab; a::b:1, or ab . a.b vif these two equations are multiplied
. .1 efi— ,. . H , 1 ‘ . -.

 

* Yule-~Intr3duction to the Theory of Statistigs, n. 31.

201

together {in}: A___ B x ab = 152]) thc‘resul-t will be- (AB) (.15) =
.1,Ba.b _ T‘ - ' . ' .

N2
In the sane way that under complete independence it has been
shown that AB: A. B. and ab = 2;; ,-it can be shown that Ab - ALE and
iv N ‘ ' N
a3 = a.B‘ If these last two are multiplied together, the result will
iii. .

be (Ab). (a3) - A B.a.b

”TE—”t

we have, therefore, in the case of complete independence.
two statements,

(AB) (ab) = A.B.a.b.

< 1; >

’Ab - aB— : '.a.b.

A~BN

Therefore A39 ab = Ab;éB. since each is equal to the same value. In
the case oflpositive association AB.a .fAb,33, and in negative associa-
tidn AB. ab «b aB. If one has all 0 the second—order frequencies,
this third criterion is perhaps the simplest one to test the associa-
tion, since the cross multiplication of diagonal values gives the re—
sult almost at a glance.

One needs to keep carefully in mind just what the ass.ociar '
tion analySis by means of the percontages stated in these criteria
'actually means.* It is not concerned with the fact that some A's are
'B's, but with the fact that the A's which are also 3‘8 (AB'S) exceeds

the number which would have been expected if A and B had been independ-
ent. It-is in reality a comparison between the number of, say, AB'S
actually found in a cross classification, with the number theoretic—
ally exnected to be found in that category if comrlete independence

existed. If we assume, for a moment, a cross classification which
does not show complete independence, it would be found that AB = A.B
' N

The number of '“'s to be expected under. complete independence, which
is usually deSignated as (AB)o, would be equal to the fraction“ A. B ,
(Since(330):A:iB:N) and a comparison of the value (AB)o with the
actual value of AB would indicate the nature of the association. This
is what is really accomplished in the criterion which tests whether
A3 = A B.

_~__

N

~ Likewise when A and B are said to be negatively associated
or disaSsociated, it is not meant that no A‘s are B's but that the
number of A's which are B‘s (AB's) is less than (AB)o, the‘number to
as ex ected if A and B had been i;1dependent. Aseociation cannot be
inferred from the mere fact that some A's are B's however great that
proportion.

 

*Yule,~- Introduction to Theory of Statistics, 3. 29.

' These three criteria suggest the way in which association"
may be tested by comoarisons'of percentages or proportions and also
sets blished fermulas which enable one to tell whether association is
oositive or negative. , One becomes much better acquainted with the
de.te. by analyzing such nercentages than by comnuting at once some
ainereitly final measure, or coefficient which, too often is the
o;11y part of the analysis to receive any attention. uOT'SVGI‘. a need
h~s oeen felt for a summary measure of the aesocie.tion exhibited in

see classification tables._ Several have been proposed and have
fee: the subject of disagreement and controversy. The coefficient
:0 osed by Yule, which he states may not be the most advantageous
in all cases; is based on the difference in the cross multiplica-
tions of the second order frequencies of a table divided by their
sum; It.i8 as follows: Q‘ (AB)§ab) - (Ab) (531.
' ‘ ( J)(a0) + (~b) (as)
It has been shown above that if complete independence existS
(AB)(ab) 3 (Ab) (a3); so that the numer.ator of this fraction in this
event is necessarily O. Th1: the coefficient of association in our

illustration would be 36 x a; $9 x.ag = 5‘ In the casa of 29? feet
6 x 9 x 9 '

associrtion where all A's were B's and all a's b's. the second part

of both numerator and denominator would disappear completely, leavi:.g

. _031Y %é§%%§§% or 1. the coefficient of perfect positive association.
ab ' ‘

In Case all A's were b's and ell P's, 3's. then the first nart of
both numerator and denominator would disappear and the result would
be —1, perfect negative association. The values of 1 and -1. may.
however, be secured when a relation of less than complete aseocia-
tion occurs. One example. will suffice to illuStrate this. If all
A's were B's leaving Ab - 0, out some of the a's were B's and some
b's then the last half of the numerator and denominator would die—
appear and a measure of one s ill result.

‘ The-foregoing analysis of a hyjothetical case where the
classification is of attributes rather than variables. illustrates
that.association is a method of analysis available where quantita-“
tive measures are not possible. Ordinary correlation ans-lysis does'
not lend itself to such data. For these situations, if no others.
the research worker in agricultural economics should be familiar
with association analysis. It is'esnecially useful in studies of
sociological data. Economic date. however, are so largely ex—
pressed in quantitative form, that a method which is to be widely
used .1ust have an apnlicetion to statistics of variables as well
he of :ttiihutes. That this is tLuc of the method of association
will a.)peor in actual illustrations.

The fundament'l ~111ciwles of comolete independence
havingfiflunestzblishcd on e.hyvothotical case, it is now possible
to take an actual set of data and a1alvze it by methods of associa-
tion. After this has been done. the analysis will be extended into
the field of partial assocL tion 111 tliis presentation it will he
. the aim. wherever possi .le, to give the analogous results o'Dtainec
- when correlation analysis is apolied to the same data. . The dat

203

selected for this illustration are quantitative measurements which re-
late to eighty;tw0'counties in the state of Minnesota. Three Veri-
ables have been chosen, the average value of land and buildings in each
county as reported by.the 1925 agriCultural census; an index of produc-
tivity which was constructed from census data on crop yields; and the
percentage of land in woodland reported by the census. The value of
land and buildings varied from-$25;09 to $149.97, with an average of
$78.61 the productivity index from 12,91 to 26.73 with an average of
19.50; and the percentage‘in woodland from 0.5 per cent to 58,0, with'
an average of 17.1 per cent. If these data were to be correlated they,
might be treated as unarouped series, or by the use of class intervals
for each variable, they might be reduced to'a double frequency correla-
tion‘table, As the size of class interval was increased the number
of classes would decrease,and if the process were carried far enough
it would be possible to condense each series into two class intervals_
and have a double cross classification table similar to the one used
in illustrating the fundamental principles of association. Any value
can be selected as the dividing point between the_two classes of'any
series, but "average and above" and "below average" are frequently
chosen, and will be used.in the following illustrations. Let the
classification average or above in value of land and buildings be
designated by A, below average value by a; average'or above in the
productivity index,B, below average productivity, b; average or

above in percentage of woodland, C, below average percentage, C.
Following is the double cross classification of the value of land

and buildings with the pneductivihy indexes. ‘

- ' Product ivity_, Index
Vfilue of land :

 

 

 

' and building : Below Average: Average and above: Total . :
Average and :Ab ' . :AB ' ‘:A ; =
above ' 2 ' 12, .: 30 ' : ‘ 42 :
Below -1' :ab 1 . :aB ‘ . " :a :
average ' . z 25 : 15 : -4O . :

‘ :b :B i ‘ . ‘ ' :N :
-Total :‘ 37 2‘ 45 : 82 I

 

Where all second-order frequencies are available, probably
the easiest test for association would be to find out whether AB.ab
is equal to, greater'than, or less than Ab.aB. 'In this case we have
30 x 25:) 12 x 15, indicating positive aesociatiOn between value of
land and buildings and productivity.

- ' Although this table could be analyzed either on the basis'
of rows or columns, the question to be answered is assumed to be the
relationship which productivity bears to value of land and buildings,

as'in the following: ,

1

Percentage of counties above average productivity among those abovev

average value: fig a go . 71.4.
A 42

204 \

Percentage of counties below average productivity among those below
average value: ab = §§ 3 52.5,
a 40 ‘
Percentage of counties below average productivity among those above'
average value: Ah 3 12 - 28 5
A 42
Percentage of counties above average productivity among those below
average value: fig . lg . 37 5
a 40

These percentages seem to point to a considerable degree
of positive association. If complete independence existed, éfi - g3,
A a
but in this case the corresponding values are 71. 4) 37. 5. Similar
results are obtained if the other tests of complete independence are
examined. *

The coefficient of association computed for these two vari-
ables by the formula AB. ab-Ab. a3 results in a value of .612. L co-
AB. abtAb. a3
efficient of correlation computed for the so .me variables treated as
ungrouped series, resulted in r a 1.457

The following associates the percentage of woodland in each
of these eighty—two counties average value of land and buildings:

Percentage of Woodland

 

 

 

Value of Land : Below : Average and : :
and Buildings : Awerage : Above : Total :
Average :Ac :AC :1 :
apd above : 35 : 6 : 42 :
Below :ac . :30 :a :
Aggrage : 12 : . 28 : 4O :

:c :C :N' :
Total : 48 : 34 : 82

 

* When one is dealing with a sample in which the conditions of simple
sampling are met, a question will arise at this point as to how
great a divergence between two percentages, such as gg and a3,

A a

‘ could be accounted for merely by fluctuations of sampling. In this
event, perhaps the best test would be to compare the observed differ-
encebetween the two percentages with the standard error of the per—
centage difference resulting from a solution of the equatione. . +W1QK

(Yule p. 269. ) Unless the observed differen’g‘e n n,

exceed some three times the standard error, it is possible for the
,association to have arisen solely by the chances of simple sampling;
Since each observation of the data of the illustration may have re—
sulted from a different set of forces, the theory of sampling is not
applicable; but if it were, the standard error of sampling would be
10.4, to be compared with an observed difference of 33.9 per cent.

205

- Analyzing this Situation as with the previous variable, we
can see at a glance that negative association is present: or ‘
AC. acZLAc. aC

6. 12\<36. 28.

The percentages which may be examined for aesocintion are:

Percentage of counties below average in woodland among those above
average value: Ac = 86 a 85. 7 ~

A
Percentage above average in woodland among those below average value:
=88: 70..o
a 40

-Percentage above average in woodland among th_03e above average value:
A0 a 6 -= 14 3
A 42

. Tercentage. below average _in woodland among t.hose below average value:

ac :fE: 30..O . .

a 40

' ' These percentages, indicate a considerable degree of negative
"association. ' The coefficient of association in this case computed by
by the formula previously used is--.867; The coefficient of correlar
tion between these two variables computed from an ungrouped series of
the eighty-two counties is - .638.

If the analysis had been carried to this point by correla-
tion methods, the next step would doubtless be to aeoertain the rela—
tionship between productivity and percentage of woodland.‘ The -same
step may be taken by aSSociation analysis. A cross-classification of
these two variables results in the following table:'

Percentage of Woodland

Productivity Below : Average and

 

 

 

Index : Average : Above : Total :
"AVerage and :Bc' ‘~ 1 :BC' 7 ' éB :
apnea ' : 26 t " 19' : 45 :
Below' ‘:bc- ' :bC ' :b 3
lNerage : 22 : =15' : 87 :
:c :C :N :

”Total. .: in, A . 48 : 34 :. 82 :

 

we- see that BC. be is slightly larger than Bo. bC, indicating
a positive sign for the coefficient of association.‘ The percentages
to be examined are: . - 5' " '-

'Percéntage of counties above- average woodland among those above average
productivity: BC: 19 = 42 2
' B 45

'Percentage of counties below average woodland _among those below average
productivity: bc : 22 :' 59 5 -
. 1 b-

.'.,. -¢- ‘ .., ................i .4 --.‘». ., .. - u m...,... ..

206

Percentagewof'counties below average woodland_among those above average
productivity: 39.. - __2_6= 57‘s,
3 45
Percentage of counties above average woodland among those below average
productivity: b__C_-l__5_ a 40 5'
b 37

If the tests for complete independence 3g - 29, gag g9 . pg
3' b B b
are applied it develops that while 42. 2 #=40. 5, and 57. 8 %r59. 5, yet
the difference is slight. The number of cases to be expected under
absolute independence, (EC)° (bC)o and (bc)o, is in any case so nearly
equal to the actual frequencies that the association must be neglig-
ible. For instance (BC)0 _ 3.0. which is 18.7, while the number of
N

Ed's actually found was_19. The coefficient of association is found
to be +.O34. The coefficient of correlation between these two vari-
ables in the ungrouped series for eighty-two counties, is 1.183-

Partial Three simple associations, one exhibiting
Association. positive association. one negative, and one

with practically no association, have been

tested out by the methods of association

analysis ' Values for the Coefficient of
correlation in the same cases (computed for ungrouped series) have
been cited for comparison. To continue the comparison, methods of
partial association must be taken up along side of partial correla-
tion. A partial correlation of these variables gives r AB.C a
value of c...758 and r AB.b —.825. which shows that the net correlar
tion between A and B, is increased positively When the effect of C
is eliminated, and the net correlation between A and o is increased
negatively when the effect of B is eliminated. (A, B, and C are
used here as subscripts of the r, without distinction between cap—
ital and small letters.)

A partial association analysis can be carried out with these
three factors which will result in substantially the same conclusions.
. Such an analysis necessitates a triple cross classification, as in
the following table:

 

value of land and Productive‘Index'

 

 

 

 

 

 

buildings, and per— : Below : Average and :
contage of woodland : Average : above : Total :
fiveragg value and above: A : : 3
. Average woodland and Echo - ‘lsc : 7' -‘AC ‘
above : #l : 5 : 6 z
' = :Abc 4 :ABc ~ :Ac :
‘Below average woodland : ll : 425 : 36 :
- :Ab ‘ ' :AB ' :A :
Total : - 1g__v : 30 : 42 :

 

207

 

 

Value of land and : Productive Index

 

 

 

 

 

 

 

 

 

buildings, and per- - Below = Ayerage and i :
centaaes of woodland : average 1 above : Total :
W41 : W
: bC :aBC :ao :
Average woodland and : : : :
above : 14 : 14 : 28 :
:abe :aBe‘ :ae =
Below average woodland: 11 : 1 : 12 :
tab :aB :a :a
Total : 25 : 15 : 4O :
All values : -: : :
Average woodland and : . S : :
above :b0 15 :BC 19 :C 34 :
Below average woodland:be 22 :3c 26 :c 48 :
Total :b 37 :B 45 :N' 82 :

 

The complete tally is made in the eight squares AbC, Abe, ABC,
Abe, abC, abc, aBC, and aBe. The remaining compartments are merely
totals derived from these Combinations. For example, the total of a?’
ABC rABC = AC, and of AbC-rAbe = Ab.

. ?aria1 association will attempt an analysis of the associap
tion of the A and B factors within the universe of C's, the counties
above average in the percentage of woodland. In other words, the
association of value of building and land with productivity will be
examined when the effect of the percentage in woodland has been elimr
inatfid. From the classification we can observe the proportion of 3's
among the A‘s in those instances where, with re spect to average per—
centage of woodland, no variation exists. » The number of counties
which are above average in all three of the factors under considera-
tion, designated by ABC, is 5. If no asSoeiation of 1's and 3‘s ,5
exists in the universe of 0's, the following preportion woould hold:
ABC:AC = 30:0. This is, within the universe of counties above ay—
erage percentage of woodland, the proportion_of counties above aver—
age productivity among those above average value will be equal to the
proportion of counties above average productivity in the universe
under consideration (that of the C's). This equality will hold in
case>there is absolute lack of association, so that a test of whether
IQACQQBCZ will indicate whether the partial association is posi—

tive, negative, or absent. The value of the fraction §AC2§PCZ is

208

equal to the number of cases to be expected. (ABC)O. under conditions
of abaclute independence. In this case, the theoretical number is
'3 and the actual 5, so that there is evidence or positive associa»
tion of the A and 3 factors within the universe of 0's. The most
3convenient comparisons to be made from the proportion ABC: AC::BC:C
are as follows:
ABJIBC; and ABCEAC.)
iAC ‘ C BC C ) ~
It is also well, in addition to making comparisons with the whole un~
:iverse of C's, to compare the proportion of the B's among the .i's in
the universe of 0'3 with the proportion of B's among the a's in the
:same universe, Ito see -if A30: agg, Iand similarly the proportion of
' AC aC
. A's among the B's in the universe of C's with the proportion of A's
'4 unong the b's in the sane universe ABC = Abe.
‘ . BC ”“ b0 .7

The association could be tested by any of these formulas,
-Ibut the two of the latter type, for the reasons given under simple
. association, are sometimes Slightly more reliable. Ii we determine
.whether ABJ; (QC. the result iis 5>14I, again indicating positive
:relationship of the factors A _and B nnder  the analysis for partial
. association. 4

In order to test out the degree of the partial association
of value of land and buildinrs ,:with productivity when the effect of
percentage of woodland is not operating, a seriesIOf percentages
’similar to the following may be examined: . -

For the entire material:

Percentage of counties above average value of land and buildings
W” 82 , I,
Percentage of thOSe above average value amoIng those ahove average
pioductivity: gs = 30 = 55-
' B 45
For those above average woodland:

Percentage above average Value: AC 3 6:13
C - 34 "
Percentage of those above average value Iamong those above average
productivity: ABC . é a 26I
BC 19 I
‘ For those below average woodland:

 Percentage above average value: Ac 3 86: 75:
c 48 '
Percentage of those above average value among those above average
productivity: ABc . 25 . 95
Be 26

.209

When only those counties are considered.which are above average~per-
centage of woodland. the association of A and B is shown in less
degree than for the material as a whole.- The association of A and

B however, is considerably more pronounced in the counties where the
average percentage of woodland is below average, than in the case of
the material as a whole. This result would Seem to he in accord with
T A3 - #2457 and r AB.C.b-+.758 which accomplishes elimination o? the
.0 factor from the series as a whole.

Similar analysis oculd he carried on to Show the association
of value of land and buildings with percentage of woodland when the
effect of productivity was eliminated from the consideration. The
statement of the proportion for the association of~A's and C's in the
universe of B's which would he expected in the case of §$801Ute indef
pendence is ABC:AB::CB:5. One test of association is whether
mac 5 (ABHCB) ', 1' '

> B'
In this case we have 5(30 3: 19 , or 5(13. indicating mesa-
’ ' i 45 . ' '
tive association in the partial analysis. Othei forms of comparison
which the above proportion suggests are $39} with g3, and ago with as.
_ AB ' ' :3 cs - "B
Here also. for reaSons given under simple aseociation c6m§h
parisons of the type égg with Egg showing a camparison of the prOPOr‘
. , ' 33' a3 ‘ - 4
tion of 0‘s among the Als in the universe 5r B's with the C's among
the_a's in the same universe; might be somewhat better. This.compari-
son §_Z;l$, shpws evidence of negative association in the partial an-
' 30_ 15
nlysis.

.Inganalyzin: the degree of partial asSOciation between A and _
C, without Variations in B. the following‘percentages should be examinedg

' ' t
For the entire material:

Percentage of‘counties-above average value: g ="51.
Percentage ahote.average value among those which were below average
woodland: fig;g @Q»: 75. " \ ' ' ’
c .5 48“ ' '

04.

 

* That ABC is‘the proportion of C's among the A's in the universe of B‘s
, CB 1 ' . f '~' .' . ..
may be seen to he analogous to the former illustration of partial
association if it is written CAB. No significance attaches to the
order of the symbols in any class designated hut the proportions to
be studied sometimes appear more clear ifwthe'order is altered. CAB
' _‘ n , p ‘ a . Ki—
. _‘ , _ ,. . ,

seems to give a clearer statement of the proportion of C‘s among the
A's within the universe of B's, than does.ABC.
' AB

. w«ms

.. --.

'\

lbr thosa.above average productivity:

Percentage\above'average value: £3 = §Q i 66
B 45/

Percentage of those above average value among those which*were below

average WOOdland...AB_9_4-25 n 96 .
Be 26

For those below average productivity:

Percentage above average value: $3 = lg = 32
b 37
Percentage of those above average value among those which were below
average percentage of woodland: Abe g ll =
be 22

Here the negative association between counties above average
value and those below average percentage of woodland is tested out with
the B factor, or productivity index, held constant; In the group where
only those above average productivity are considered, there is evidence
of greater association between counties with the high values of land
and buildings and low percentages of woodland, than is evidenced in the
material as a whole. In the separate group of those counties below
average productivity there is no evidenCe of increased association. In
fact it is scarcely as high as in the material as a whole. When the
data are all handled together by the method of partial correlation
applied to the ungrouped data, the r AC. B; is found to be equal to
-.825, that is, the net relationship between‘value of land and buildi;'
ings and percentage of woodland is somewhat increased by eliminating
the effect of productivity which is in agreement with the association
analysis in the universe of B's. The r ACv it will be remembered,
was -.638.

By the methods of simple association, it'has therefore been
possible to analyze the relationships between each two of three diff-
erent variables with substantially the same conclusions as those which
would have been found by the gross coefficients of correlation. BY
methods of partial association, something has been demonstrated of
the nature and degree of the net r gégtionship between each of two of
the variables when the other/m eliminated. The elimination
was accomplished by means of a cross classification which madeit pos—
sible to study the relationships in two groups, in each of which the
third variable was constant (1. 6. either above or below average).
Although the conclusions are necessarily less specific than those of
partial correlation, there seems to be general agreement between the
conclusions from the two methods. Association analysis, however,
has nothing to offer in lieu of the regression equation or multiple R

~ of the correlation analysis.

~The methods of association provide a means of analyzing
data by the simple and easily understood methods of cross classifica-
tion, and£a comparison of the percentages found in different combina-

211

tions, with those which would have been expected if no tendency toward
association had been present, This type of aSSociation analysis, how—
ever, permits of only two group headings for each variable, and when
this is done, a considerable amount of detail is lost which can be
taken into account in correlation analysis. In correlation analysis,
all observations are measured by the product of their deviations from
the two means and related to the square of the deviation from their
respective means through the two standard deviations, as r: E(x.i2r
NJXKy
In association, all cases which fall in a given crose claSSification
group have equal consideration'regardless of whether they are extreme
measures or barely of a size to come within the classification. In
general, this is an argument against association analysis. However,
it eliminates the difficulty sometimes experienced in correlation an-
alysis where one or two extreme cases produce a coefficient of signif—
icant size when no general relationship exists at all. '

Association analysis may be accomplished in much less time
than correlation analysis and may often be'employed as a preliminary
step in considering the variables to be used in correlation analysis,
even if it is not used as the final method, In fact, when scatter
diagrams are made, association analysis may proceed at once from a
count of the cases in each quadrant, or on any Other basis of class-
ification desired. In the analytical study of data, such as is in-
volved in association analysis, the research worker is likely to be—
come more thoroughly familiar with the data than when the analysis
is made along the more routine methods of correlation.

Finally, the results of association analysis, which are
expressed chiefly in the form of percentages or proportions, may be
presented successfully to those who are unfamiliar with the technical
phases of correlation, and to whom the results of a cOrrelation anal-
ysis would mean mothing. '

Measures 0f Simple measures of association-apply only
922$i££§2§1.. to data in a 2 x 2 fold arrangement, where

the classification has been made accord—

ing to two characteristics, each of which

is divided into two headings. The co—
efficient of contingency has been devised for the analysis of data
in manifold classifications which do not permit of numerical ex-
pression. The method is apolied in Table I to an analysis-of the
association which exists between geographic location and the rate
of interest on farm mortgages. These data are reported in the 1920
census_for nine geographic divisions and seventeen interest classes
of varying 81293. In order to permit the table to be more easiLL
handled and at the same time allow a considerable amount of detail,=,
the class intervals of interest rates have been condensedrinto-nine
relasses. The number of farms reported has been expressed in hun-
dreds. '

4 - .212»: 4

Table 1;'~ number of farms in the united States on which rate of
' interest on mortgage debt nearrepcrted in the 1920
census, classified according to rate of interest and
gGOgraphic divisionsr*' -

.-(In hundreds)-

 

: Interest rates
:Less: 4.0: 5,0:.5.5: 6.0: 6.5: 7.0: 8.0:10.0:

 

 

 

‘ :than: to:: to: to: to: to: to: to:anh>:Total
Geographic. : 4.0: 4.9: 5.4: 5,9: 6.4: 6.9: 7.9: 9.9:0ver:.
DiviS' . : :‘ : :' ;,. :» : : : :

. 1°“ 5A1.:"‘2:,A35A43‘.‘5:‘6,,A7fi‘ef’s :

he".§ng1_and._ ____I;L:_ 1 : , 8:150:,35:.-274: s; ’7: 9: 1: 488
Middle Atlantic ' 32: 8 : 64:'440: 74: 518: _ 1: 1: l: 1: 1108
E.N.Centra1 B3: 13 : 134: 537: 200:1422: 42: 3 8: 30: 3: 2709
N.N. Central '34: 3 : 5 : 437: 363:1077: 11": 280: 278: 27: 2628
South Atlantic B5: 2 : 9: 39: 45: 565: 5: 60: 247: 27: 999
E.So.Central 135: 1: 48: 31: 41: 586: 4‘: 25: 315.: 51: 1052
West So.Centra1 B7: 15 : 10: 70: ~63: 238: 20: 114:.477: 273: 1280
Mountain 38: O : : ,19: 36: 135: _15: 109: 294: 113: 725
Pacific 139: 1: 6: 33: 33: 236: 24: 255: 106:. 9: 703

Total : 4 ::

 

298:1756: 890:5051: 227:ll79:1752:'505:11702

Correlation analysis of these data would be impossible, since a
quantitative statement of geographic location is not available.' The
quantitative measurements of interest rates as reported by the census
also contain class intervals of varying sizes with unsatisfactory
midpoints, which make this variable also poorly adapted to correla-
tion methods. Contingency analysis does not.assume equal class in-
tervals or consider that all frequencies within a class fall at the
mid—value. ' ‘

The single meesxre of association, such as that which exists be—
tween geographic location and interest rates in this case has its
basis, as in simple association, in the difference between the number
of cases found in each compartment of the classification and the numy
her to be expected in that compartment if no association at all were
present.

The system of symobols need in the table is self-explanatory.
The number of farm mortgages which carried an interest rate of "4 and
under 5 per cent" located in the East North Central states would be
designated by a combination of symbols, Ag 33. ' The frequency of
1 £2 33,'1n«tholtab1é.is 134.

 

*Abridged table from 14th Census - Vol. V p. 491

213

. one column of the Table, A3, will be studied in detail in
order to illustrate the principles involved. A total of 1,756 farms
(OO omitted) were reported at the interest rate of 5 and under 5% per
cent.lr,If these 1,756 cases had been distributed among the different
geographical division in the proportion-which the total number reported
for each geographic diVisiomywas of the total number for the United
States, we should have the distribution expected under a c0ndition.of
absolute independence. The expected frequency or independence value
for any classification will be indicated by the usual designation with
the addition of parentheses, and a zero subscript.' Under absolute in—
dependence the value for (A5 311° could be-found on the basis of pro~
portion existing in the totals A2 Bl>o : 1756::488:11702, or if one
prefers, it may be thought of as A3 Bl)O : 488::1756:11708. The latter
statement is to the effect that if no relationship existed, the number
at farms with interest rates of~5 and under 5%4percenttin New England
would bear the same relation to the total number of farms mortgaged in
New England that the total number reporting aerate of 5 and under 5% per
cent did to the total number in the whole country. The arithmetic is
the same whichever statement is taken;: 'The independence value for
(15 Bl)o is 1756 x 488; and (A3 Eg)§ja;l756 x 1108_‘;H

11702 : , _ _ .' ~11702 . .
Independence;values corresponding to the frequencies of each
compartment in the original-table, have been similarly computed, and
are shown in Table L1,; -- g - - ' .
Table II.s- Independence values for the Frequencies
of the Preceding Table.;

 

 

 

   

 

 

 

: - :Interest rates : :

4:685:10: 5.0: 5.5: 6.0: 6.5: 7.0: 8.0:1020: :

Ge°€TaPhiC :than: to: to: to: to: to: to!- to: aihzTotal:

DiViSions' : 4~09 4:9: 5.4: 5:9:-6.4: 6.9: 7.9:‘9;9:over: :

New England . z 2':, 12: 73: 37: 211: .9: 49:.“78: 21: 487:
Middle Atlantic : 4 : 28: 166: 84: 478: 21: 112: 166: 48: 1107:.

E- N. Central : 10 : 69: 407: 206:1169: 53: 373:" 406: 117: 2710,:

W. N. Central' : 10:: 67: 394: 200:1134: 51:”265: 393: 113: 2527:

South Atlantic : 4 : 25: 150: '76: 431: 9: 101: 150:"-",43:‘ '999:
E.South Central : .4‘:x 27:.159:' 81: 458: 21: 107: I59: 46: loszkg:
W. South Central_ : :5 : ’33:;192: ’97: 552: 25: 12%: 192:' 55: l“ .h

Mountain : :8*:: 18: 109: 55: 313: ”14:” 73: 109: >_; 5:

Pacific ..: :3: 18: 105: 53: 3031-714: 71: 105:; $01“. 702:

' a :‘1 : :‘” :L.: :_ g . j:, :75 ,
Totals* : :45 .:'297:1755,::'.§89;. 49;! Tin 53: 504:1;599;

* Since this tablegiaighr9~d£stf§bfitioulof the actual valuee'bi:
the preceding table, the tefihls”should theoretically agree:‘
The small differences which occur result tram rounding of! i .

the independeneefivélues to the nearest whole number.

The differences between the actual frequencies and the
independence values for colmmmffie" ught together in Table.
III in order that the differences may be more learly seen. The
difference column shows larger numbers than would be expected in
the New England group, the Middle Atlantic, the East North Central,
and the West North Central groups. and the opposite in the five
southern and western groups, under a condition of absolute indepen—
dence.

Table III — Independence values and actual frequencies
compared for Column Three.

Farm mortgages reported at 5
and under 5% per cent.

 

 

Actual : Independence :
No. : Value : Difference
New England :A331 150 :(AgBi)o 73 =d31 #77
Middle Atlantic :Ang 440 :(A332)0 166 :d3 +274
E. N. Central :AsBz 537 :(A333)o 407 :dss +130
W. N. Central : 437 : 394 : 443
South Atlantic : 39 : 150 : —111
E. So. Central : 31 : 159 : ~138
W. So. Central : 7O : 192 : —122
Mountain : 19 z 109 : - 90
Pac ific : 33 : 105 : - 72
: 1756 : 1755 :

 

The corresponding comparison for colunn 8, shown in Table II,
indicate a reversal of the foregoing situation for the higher interest
rates.

Table IV. Corresponding analysis for Column Eight.

Farm mortgages reported at 8
and under 10 per cent.

 

 

Actual : Independence :
No. : Value : Difference.
New England :A331 9 :(A331)o 73 =d81 ‘54
Middle Atlantic :Ang 1 :(Aeflz)o 166 zdaz -155
E. N. Central :A833 30 :(1833)o 406 :aes -375
W. N. ’ tral L 275. x , 393 i -120
South antic : 247 : 150 : T97
‘ E. So. Central : 315 : 159 : 4156
W. So. Central ,: 477 ,: 193 : +285
Mountain : 294 : 109 : +185
Pacific : 106 : 105 : + 1
A 1752 : 1753 :

 

215

The coefficient which measures the relationship between the two
variables is based upon all such differences which exist between the actual
number and the number eXpected. Since the differences (d's) are both
positive and negative, these values are squared in order to eliminate al—
gebraic signs. Each of these squared differences is then expressed as a.
ratio of the corre3ponding independence Valu . If we think of the differ—
ence between -

AiBi- (A1Bi)o - c111
AlBg- (A132)° = dlg
A233- (A333 )0 = (123

and of each one in a. similar fashion, we may conceive of the series con-
tinuing indefinitely until we have AmBn) - (Amlan)o -.- dmn' In these gen-
eral terms, the arithmetic steps described can be summarized in the follow- '
ing formula for what is known as the square contingency, designated by
Me as x3 . _. ,_

j><L QEI/fii {trufg )

This expressionmis ralcways positive, because it is a sum'cf
squares. If there is a condition of absolute independence, each actual
value will be equal to the expected value, so that each 6. will be zero,
and consequently 237111 be zero. The coefficient of contingency on
this basis has been defined by the following:

C:Z<2fi.
N+)<

This coefficient is zero if the characters A and B are inde-

pendent,- and approaches more and more nearly to unity ast increases.*
In general no signshould be attached to the root, as the coefficient

shows merely whether the two characters are independent or aesociated in
some degree. In some cases, however, a conventional sign, which is deter-
mined by inspection of the original cross claSsification, is attached to
the coefficient. -

 

In practice, a formula for the square contingency (X2) may be
used which eliminate the computation of the differences between actual and

theoretical values. The formula by which the method has been explained may
be transformed by the following steps to a more advantageous form:

2 2
)< “Elf‘ffimiw O } V y
X 2 =€{%§M§nb_? (in which 6. is replaced by its equivar-
0 lent, Aran' ~ (Ankh)-
X 2 =E{ 3 2- 2 2 (expanding numerator)
fr' 2 ‘ 2 . .
" - 2 ‘ In 6. 1' all f
X 5 -‘Wm “use + tubs}; .V<.1n::r::::fz(2;3n) we?
(AmBn)o~ : N
2 2
X -5 mini - N
(AmBnh

*Y‘ule - Introduction of Theory of Statistics, 13. 65

 

in; \~ . .

If the value of the summation cf-the quantity in ‘rachets he
designated by S, and 2 be written as S—N;-the formula for the co-
efficient of contingency may be changed to a new form also:

2 ' A
Ifc-N—x__3_, w_,~‘._
' *‘)( substitutin€)(2 : S—N
0. {$41) '
N +~ S-N

xx”; .1359: NML \35 iii .

C S-N
8 (AMBnI

“I

These last formulas

were the one actually used in the comnuthtion of the square contingency
_ and the coefficient of contingency for the data on interest rate on farm
mortgages and réopraphical location, since they involve more simple com- -
.‘uitation. This means that the differences (d's) we.re not actually com-
puted but that the actual nmnbers of ferns in each interval were squaredr
divided by the independence value for that class, the resulting values
~added, and the value of N subtzacted from this amount. The resulting
figure was the square contingency. and the coefficient formed from it

was .568 which indicated a considerable degree of association between
farm mortgage interest rate and § mogrn~hic location.

The order of arran:ement of the headings in a Contingency

stable has no effect upon the value of the coefficient. In this case I
the interest rates have naturally been' arranged in a numerical sequence
and the order of reoeraphic divisions is that used in the Census. If,
however. a conventional sign for the contin~ency coefficient is desired
it is necessary to inspect the ori2inal ta.‘le with some definite sequence
.of .characters in mind. Thus. if :eozzaphical location is considered as
' proceeding. from East to WeSt. positive associre ,tion is pre sent between

it and an increase in interest rate andI the CoefficienItI of conting;ency
may be stated as {~568.

- The formulas- for measures of continb ucy already given are
those of Ehhs. Pearons has used 9. measure for the mean square con-
tingency which he denotes bygé. 3 This is in reality >32 -but Ithé

1,, N
character is often used that it seems advisafile to connect it with

the formulas and termin610;:y of Ithi-s nrtich It would thus be
expressed, . . . ‘.;. .

 

217"

The goefficient of contingency ordinarily stated in connection with
is . . ~ ' -' '

,

   

This can readily'be derived from Yulels formula C .” 2' by
dividing both numerator and denominator by N,_ and substituting ¢ 3

for g2 ,.

N '

‘Many objections have been raised to the Coefficient of_con-V
tingency. and much theoretical discussion has followed with little
agreement; One'bbjection, however. is of sufficient practiCalzim-
portance in the interpretation of coefficients of contingency.that
it seems advisable to mention it briefly. This is the caution that
two coefficients calculated from the Same data but dn-different
systems of claSsification are not new}: with ea‘fhother“ . If
a coefficient of contingency had been computed for the_same data.ae
the above but with different grouping83‘say six intervals of interest"
rates and six geographical groups,-the coefficient for this 6 x 6 .
fold table would have tended to be less than the one compared for the
9 x 9 table.* 5' .,

The coefficient of contingency‘of goegraphic location and
interest rate on farm mortgages which was found to be .568 on a
9.x 9- fold classification, dould not have exceed a value Of .943
had the association been perfect. More groups would have raised
iit; fewer would have lowered it, The greatest possible values are
'so low for Smaller tables that Yule considers it wise to restrict
the use of the coefficient of Contingency to tables 5 x 5 or larger.
Yule gives e table 'of the limiting values of the. coefficient of '
contingency with different numbers of groupe.**

. The coefficient of contingency_is an imperfect measure of '
association between two characteristics. It use is not recommended
for series where both characters permit of satisfactory quantitative
measurement. In such cases, the-coefficient of correlation or core

. relation ratio are more satisfactory; But; where one or both of the ‘
characters permit of only qualitative observations. or the quantitae
tive measurements are unsatisfactory for.correlation, the coefficient
of contingency offers a-means of measurement of association between '
the characters, ‘- ' « ’

It will be interesting to measure the association present
in the above contingency table. if it-is reducedvin different ways
to'3,X'2- fold form and the elementary analysis of simple associa-

-tion applied. Yule's statement in'this regard is as follows: “it

 

* See Yule for a demonstration of.thiS.
** p. 55'

Q18

then becomes possible to trace the association between any one or
more of the A's andsny one or more of the B 9, either in the universe
at large or in universes limited by the omission of one or more of
the A's, of the B's, or of both."* If the division is made on the
basis of east of the Mississippi and west of it, and under and over

6 per cent. the coefficient of simple association is only 4.1944

The following table shows a different geographical division:

Interest rate.

 

 

 

Location : Under 6% : 6%iand above : Total 3
"West" (Mountain dnd:Ab :AB :A =
Pacific divisions) : 132 : 1296 ' tags 3
"East" (all other) :ab :a3 :a z

: 2856 : 7418 : 10274 :
:b :B :N :
Total : 2988 : 8714 : 11702 :

 

The coefficient of association is +.582, and the significant percentages

are as follows:

Percentage of mortgages 6 per cent and above among these re-
ported for the entire U. S. 3 a 8714 a 74.5,
N 11702
Percentage of mortgages 6 per cent and above among those in
the "East". fig n 7418 a 73.2,
a 10274

Percentage of mortgages 6 per cent and above among those in
the "West". Ag = 1296 . 90.0_
A 1428

. The division of the interest rate at six per cent for these
cross-classifications was entirely arbitrary. Much higher degrees
of association would doubtless be found if other interest classifica-
tions were analyzed. A large number of other associations between
geographic losetion and farm mortgage interest rates could be as—
certained by voribus rc-groupings of the detailed contingency table
into 2 x 2 fold classifications. The particular ones to be made
wOuld depend entirely on the nature of the questions to be answered
inia specific research problem. In the case of data qualitatively
expressed, the coefficient cf contingency is oftentimes a convenient
measure for a situation taken as a whole, but it should ordinarily
be used as a supplement to, rather than as a substitute for, an
analysis of association based on fundamental principles,

 

I"Yuleylntroduction to the Theory of Statistics, p. 62.

219, ‘

In this section the objective will be to outline and explain
the method of correlation analysis in terms of problems in agricultlral
economics, and indicate the significance of the-results so far as the
data analyzed are concerned. Correlation and regression coefficients
\. are to be understood in this section as descriptions of the sets of

-data analyzed. ' A following section will take up the subjects of in-
ference and prediction. = .t

Elementagy It was thought best to introduce the sub—

methods of , ject of correlation analysis by discussing first
Anal sis*' , the more elementary methéds and then-showing what

can be accomplished by more refined methods. The ‘

‘ . 'data used to illustrate this:nre hypothetical~
largely and relate to the value of the land of-SO farms located in one
quadrant of a circle enclosing a city of five thousand. The analysis is
therefore of the crossLsection type, local geographic factors being the
variables. The objective is to account for the variations in the value
of the land in these farms.~ The farms range in distance from town from
a quarter mile to twenty miles, the modal distance being 17 miles. - The
distribution is thus strongly skrewed so far as this variable is con-
' corned. The data taken for this farm include also the percentage of:
- improved land and the type of soil roughly classified as "loam", "clay",
and "sandy", these being the names commonly applied by farmers in the ‘
area. other information was obtained, but is included to a limited ex-
tent in the present analysis. For example, six of the farms were on
the shore of a small summer-resort lake. Some of the farms are upon
main-traveled improved roads and others-on crossroads.

.A method of_ana1ysis commonly followed until recently is ’
illustrated by Tables A, B and C following, namely, grouping the farms
according to one independent variable at a time and computing the avers ‘
age values of the dependent variable for these groups. It would appear
from this that all three of these factors are definitely correlated with
land values. But the results are not dependable unleSS it can be de-.-
monstrated, first, that the farms are equally improved and have the same
distribution of soil types, as.woll as road frontage and all other pos—
sible factors. at all the different distances from town; and scoond, that,
enough.ferms are ineluded in the sample to give all these different fee; '
tors a chance to average out at all distances from town. Furthermore,
such tables provide no measures of the amount of correlation. .Plotting.
the group averages as in the solid line in Figure II suggests something
a .as to the regularity of the covariations and their linearity. More
-; farms would make possible more groups and give more detailed suggestion

As to this regularity and linearity. .But there would be no definite
proof or measure of either of them.» Comparing the ratios in the last.
columns in the three tables would seem to give a measure of relative
rates of regression; For example, one would say that apparently per-.
pcentage of land improved had the greatest effect on land values and soil

 

 

* Mr. G. C. Haas of the Bureau of Agricultural’Economics_assisted Somewhat
with.this section. - . .

-:.l-

220'

type least. One could even fit a line or curve mathematically to these
group ratios and reduce the comparison to definite terms. But the con—
clusions would be valid only if the conditions of non-correlation and
sufficient number of farms above described are fulfilled. The regres-
sions for improved land may include some of the effect of distance from
town if the two are correlated, and vice versa. The early reports of
farm business surveys are replete with examples of this form of analysis.

 

 

 

 

 

 

A 3
Groups- :Average :Avorage:Ratios:Groups :Average :Average:Ratios
Percentage of:Per Cent:Ve1ue :of :Distance: Distance:Va1ue :of
Improved Land: Improved: of Land: Values: to Makt.:to Town :of :Values
: : : : : :land :
20—39.9 : 29 : 44 : 100 : O to4.9: 2.2 : 204 : 100
40.59.9 : 52 : 98 : 222 : 5 to9.9: 7.3 : 157 : 76
60.79.9 : 66 : 148 2 836 :10tol4.9: 12.5 : 133 : 65
'80-100.0 : 88 : 185 : 420 :15t019.9: 17,0 : 91 z 44
C
:Soil Types:Number:Ayorage :Ratios:
: :of :Value :of :
: :Farms :eer Acre:Va1ue :
: Loam : 19 : 175 : 100 :
:Cley : 25 : 126 : 72 :
:Sandy : 36 : 112 : 64 z

‘\

The next step taken in the progress of correlation analysis was
to compute Pearsonian coefficients of correlation and regression. The
results of such an analysis are given in A and B in Figure 1. Percent-
age of improved land shows a coefficient of correlation of .82. and
distance from town of -.77. The regression equation for percentage-
of improved land is y = 2.46X; that is, each one per cent increase of
improved land is accompanied by an increase of $2.46 in the value per
acre. Comparing the lowest and highest groups in Table A. and the'
same as represented in terms of ratios in the solid line in Figure II,
would give y . 2.39X as the equation; comparing the lowest and the next
to the highest would give y: 2. 91X as the equation. The true regres-
sien'lies between these, but is nearer the first. Our formal correla-
1ation analysis has given us two definite mathematical expreSsions of

'what we saw in Table A and Figure II. One of these expressions the

regression equation, is in terms of a straight line fitted to the data.
If currilinearity is present, if one per cent in improved land adds
less at the top than earlier, as is somewhat suggested by Figure II,
then neither the correlation coefficient nor the regression equation
.are accurate mathematical expressions of the data. The other math-
ematical- expression, the correlation coefficient, lias taken account
of the amount of scatter about the line of regression. Figure I
shows considerable of such scatter. It frequently happens that
tables such as A, B and 0 show high regularity even though the ob-
servations are very widely scattered because of other factors or er-
rors in the data. Camputing a coefficient of correlation is a much—
needed corrective for this.

SCATTER DIAGRAMS SHOWING RELATION OF LAND
IMPROVEMENT AND LOCATION TO LAND VALUES

RESULTS OF CORRELATION
BY COSS-TABULATION

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 
    

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

$ 25c r .
r- 0.82 F 6. IA 0 FIG.2
= 2.46)!
2 — 50
2 5 o Sandy36 - , 1: 5
x C/dy 25
200,— ' Loam [.9 = 500
0 .
Q . It.
g I75 ' " 5 1.50
U .l o g o
c . V f
O x X , ”
§ ISO . . I: v r z 1'00" Simple correlation V
q r: 5/3212 {mean} . m y_ s y
‘c '25 I c = 350?
t x O K /
& IOO (x t P 3 300°
K I 8 3L U: ra—o
D V [o '2 D' to ’9"..._....o
o o Is nce _.
g 75 x x ‘2 250% heldconst. {F
E x / ° % :‘Soil type
> 50 - 20° / 9" held constant
'/
IL“ 00
25 / n I50
/ X= 62. 5°/ {Inca/L) 400
I0 20 30 40 50 60 70 80 90 IOO
Farce/21093 of‘ Improved land
—$250 ' I I X=ll 6m1 {méan} 5°
0 " FIG . IB '
225 '
\x . ‘ 20 40 . so so :00
\\ Percentage of Improved land
220 < ‘ =
Q» .\.~ '
E) I 5 \ x .0
a ° 0\ o x ’. » "
K V L J, >\\x ‘. X
Q) I50 0 ‘5: D 0'
°\ ° 0 a x ' c. Y
b '25 =5] ./2/ can) A V _
Q I \< o
S o 1 " “Q 5‘0
k I O 0 ° ‘. .
o 0 00 X \
0
05) 75 x “x c \\
\
E x 0
5 5° r- —a77
y- = — 7 46‘ x " \
25_ 0 Sandy 36 00 a?
x Clay
._ loam /.9 X
I 1
2.5 5 7.5 I0 I2.5 I5 I15 20 22.5
Miles from lawn

1
a“

|
" “34%."
'v

 

 

n.

V": I .‘!‘~
4:“ 9&1 l
-.: '3

 

 

 

The equation for distance from'market is 6 I 7.46x. Thus
with each mile away from town, land values decrease $7.46 per acre.
The scatter is a little wider in this case.

The effect of soil type cannot be measured like that of
the other two variables because soil types cannot be quantitatively
expressed. Association analysis would haveXbe used for this.

A common and commendable practice in such analysis nowadays
is the construction of double—freQuency diagrams, as illustrated in A a
B of Figure II. It is Best to make these at the start before any co-
_efficients are computed. The coefficients can be computed later if they
seem to be needed, and the results added to the diagram as in Figure II.
These diagrams reveal the-amount of scatter, and strongly suggest, although
they do not demonstrate, the existence of linearity in many cases. Soil
type could have been given a separate scatter diagram: but the scatter would
have been significant in one dimension only. ‘Combining soil type with the
other variables by means of symbols reveals this scatter and some important
relationships besides, as will presently be explained.

Testing out possible inter-correlation between the three in-
dependent variablee is the next step. This can be done directly for im-
proved land and location by calculating the correlation coefficient and
regression equations for them: but a better procedure is to construct a
double-frequency scatter diagram, or perhaps a double~frequcncy correlation
table such as Table D following: '

 

 

 

 

 

'A Table D
3 8 :' t = i
: : z j : : :
; 8° - 100-° = +++++ : ++++ = +++ : ++ :
‘ ' : 4t : : :
'6‘ ‘ V’ . ‘.+++++ ’+++++ ‘+f+++ ‘
e . _ . . . . .
s ; 6° 79-9 ;. ++++ 2+++++ 2:311 1*“ 2
p . . ' I : T. 2 .
E I Z “ I f ' I I
E I ‘- Q I ; I ;
334°— 59-9 3 i +++ 3+++t film a
“a I I I I I'Hl I
3; S Q . '“E 2 5
: ‘ ‘ : : : : :
g :3}! 8&9 : : x +~f s++++f :
2 : ' g : z : :
a : a : : :T :
Q Q o — 4.9 g 5 — 9.9 :' 1044.9 1549.9 ;
3 i 3 I i :

 

Distance from Town

224

This reveals clearly the fact th€b~the percentagprofelanASInsh~
proved falls off rapidly away from the city. Figure I B shows that the..a
soil is predominantly sandy away from town and that most-of the loam soil
is nearer town. Figure I A shows that the loam soil is more largely is»
proved than the sandy clay soil. One must conclude from this that a cone
siderable part of the apparent correlation between land improvement and
land value shown in the coefficient is due to the fact that the me im-
proved land is nearer town and has better soil. The rate of regr sion
is also increased for the same reason. These same interrelationships
likewise raise the coefficient of correlation and regression rate for
distance from town and land values. A coefficient of association for
soil type and land values would be raised for the same reason. the good
soils being both nearer town and more improved. Double-frequency corresmg
lation tables like Table D would show this in more definite fonn; and .
also demonstrate how such tables can be used with qualitative classifies»
tions when scatter diagrams cannot. ’

Thus far we have only proved the existence of interrelations.
that destroy the significance of our simple measures of correlation. Wb'”
now have the task of eliminating the effects of these interrelations. The
simplest, but in many cases inadequate, method for doing this is that of
sub-sorting and cross-tabulation, illustrated by Table E, which is like
Table D except that the land values for each square are averaged and re-
duced to ratios of values according to both of the variables. The data
in square "5.0 to 9.9 miles" and "60.0 to 79.9 per cent" indicate that the
10 farms that belong in this square average $155 per core, which is 87 per
cent of the values for the first set of squares to the left. which are
taken as the base for distance from town, and 286 percent of the values
in the row at the bottom which is taken as the base for percentage of
land improved. The feature of this table by virtue of which it accomplishes
. its purpose is that the ratios in each row or column are figured on the
basis of the values in the same row or column“ Thus the $165 in the~
square mentioned is figured as a ratio of $189 in plece'of $217. In this
way the effect of decrease in percentage of land improved from $217 to
$189 accompanying the decline from 80.0 —lO0.0 to 60.0-79.9 in percentage
improved - is eliminated from the comparison. The $145 is also compared
with the $189. When there are no values next it in the-same row or
column, as for the square "lo-14.9 miles and 40.0~59.9 per cant". be
ratio can be figured. The next square to the right of this. however,’
can take this one as its base, calling its ratio 76 instead of 100, borrow-
ing this ratio from the other-squares in the same colunun In the table,
the ratios thus borrowed are enclosed in parenthesis.

 

 

 

 

 

 

 

Table 3*
: : 100 : 79 : : : :
: Class :0 217..-.r. ‘ 1'73 : : -: g“ :'
: 80-100 :8 (5)' it} (4) z (3) = (2) _-: m 2;
: a : 1 : : z
: ‘Q; 100 ‘Q; 87 :lh (76) : 60 : :
'8 : (so-79,9 :g 189 :8 165% 145* :8 114 :33 :
3 : :~« (4) :~4 (10): (16) :“‘ (8) : “1 :i
d : : : : . : : :3
e : : : : : : :
E : » : : : 76 : 52 : :
5‘: 40.59.9 : : :E? 116 :5? 89 : E? :
H ' ' : :N (4) :“‘ (13) : N z
m : : :' : : : :
o o o o I O O I
q) 0 I I I I 0 O
m_: : : .: =c> (56) : <3 3
3 : 20.39.9 : : : (2) :9. 37 : 9. :
g o o 3 o in (6) s I
a I I I I ." I I
81 : : 0.9—4.9 : 5.0—9.9 : 10.0-44.9 :* 15.0—19.9 : :
: : ‘ : : : : m x,
: : : : :' : g) a?
: Average : 100 : 83 : 76 : 56 :.§ :
1_ : : : : : ‘ 2

 

Distance from Town

_ The ratio of 52 that goes with $89 per acre was calculated by
dividing $89 by $116 and applying this ratio to the 76. In isolating the
effect of land improvement, the 265 ratio was calculated by dividing $145

by $116 and applying this to the 240 taken from the other square in the same
row. The 286 ratio for $165 is the average of the 265 and 308 ratios in

the same row. The more squares that can.be averaged to get such a base ratio,
the stabler the comparison. . ’ '

The average ratios at the bottom and to the right are simple av-
erages of the ratios in the squares. Including or omitting the ratios in
the parentheses will have no effect on the results. It is evident that in
a crude way the effects of the two variables have been separated from each
other, that in the one case distance from town has been held constant, and
L=in the other, percentage of land improved. The ratios of values with simple
"(two-variable)'correlation analyzes were 420 for percentage improved, and 44
for distance_from town. The first has been reduced to 295, and the second
to 56 (inverse correlation). The broken line in Figure II shows this graph-
ically. Land values do not rise nearly so rapidly with percentage improved
when the effect of the nearness to town of the more improved land is removed.
Also clearer indications of non—linearity appear with the higher percentage
of land improved. ' '

 

*Anerages are omitted in squares having less_than 4 farms.

.226

The crudity_of this analysis largely grows out of the fewness

‘of the observations. If enough farms were included to give stable averages
'for all the squares, the ratios would tend to be the same in all squares in
:the same rows or columns..and not as far apart as 265 and 308, or 79 and 87,
for 52 and 60. All the ratios for percentage improved arezhigh because the
~ $37 average in the lowest group is accidentally low partly as a result of
’being based on only 6 cases. Taking the next square above as a base would
'make the ratios 100,-127 and 143 for the effect of land improvement. Group—
“ing by 3's in.place of 4‘s, as in Table F, gives still different, and prob-
‘ably more dependable results, since the averages in the different rows and
‘.columns are more stable. Given cases enough, changing the number of group-
‘ings and Varying the class limits would change the ratios very little. Also
’the more cases, the more groups that can be made, and hence the more detail
‘thot can be shown. Shifting from.4 to 3 groups has covered up some of the .

:detzil of the linearity shown in Figure II.

 

 

 

 

 

Table F. ,
: : 100 Q' 85 7; 60 'E '2
e : 71 a over :‘9‘3 195 :3; . 16? :: 'é‘. 118 : E :
E : ' : (15) :. (4) _: (4) : :
8': : Q : g g
S : : :G5 100 : “J 79 : 62 : :
a: 512-70 :5- 159 :g 126 Ag 99 :kg :
a : :~« (15) : (18) : F' (7) : r4 :
M : : : : : :
o : : 1 : : :
% : z : (82) : 66 : :
3:50&less: . :0 69 :8 58 :8 :
g : : (2) :3 (4) : H (11) ~' : H :
2 : : J! : - . z ' :
3 : : .‘ : I“ : " : z
: : 10 & less : 11 to 15 : 18 & over : a, :
: : : t z : 2 :
: , 3 : : : 2 -:
: AwermgL, : 100 : 82 : 62 : 4 2

 

Distance from Town

I

As with the grouping on one independent variable, no definite ex-
pression of degree of correlation is available from such an analysis: and al-
though a line or curve could be fitted mathematically to the ratios, the re—
sults would be valid only if other possible factors affecting land values were

gnot correlated with the two variableS'taken, and if enough cases more taken so
'that even those factors not correlated will balance out on-each side of the
regression line'or curve throughout its_whole distance.
.ated with both distance from town and land improvement.
line in Figure II show how holding soil type constant will reduce the apparent
effect of land improvement on land values when the method of cross-classifica-
. tion is applied to the problem. Soil type would appear to have more effect
than distance from town in raising land values along.with percentage of land
'improved; but the averages in the squares are so unetable that one would

surely not wish to insist upon this.

Soil type is correl-
Table G and the dotted

227 ’

Table G-

 

 

 

 

 

 

 

 

 

 

 

 

1 r
x .1 o
I. II II II II I. .l I. I. I. I. I. II I. I. I. II I, I. II II I. I. Of II II IL I. II I. II I. I. II I. II II I. II II I a I. 'I II II II TI II I-
S I
t
6 mm
m 0
Em m -9 mt .
mm mom 8H «2 m mm m: mm mm 03 .24
o r
tlf.
QC on I. II «I I. '0 I. it ‘0 II II .5 3| 0- 0' 00 It I. in II I. OL mnt II In I. I. II on It Co on It I‘ on on I. 9! I. ‘0 l0 .1 to IQ‘QI II it c. I.
0 w s
r 6 Va )
f 615 d 7 6 8 9 4.) 8 4 ) \I
d—Dg) 534.. 600— \.I 8 B a aw I.\ l/\ 1( I\ 7
m 7 5 9 7 2 1 5 8 1 2 6 ctf
1 l\ 1 [Ix .I‘r /\ “ml
S .\.. . . . .
ea man may a m w 2 g E
I. I. II II I. .I I. II ‘0 II II II II II OI II II I. I. II II Cl 0' M s O I. i, I I. II II I. II II II II I. I. II II I. I. II II II II II IIrI I. II
.a
.mw e,
mum
1...? 951 96 P4( 8 .ltr l 88/\ 72/\ 866 /\ 8
C /«\ 1 (A ( fiin. H C ( 1 1 II\
MR mom 9: Msm a“ w. . mm mm “mm: .
‘0 .6 II no cc 00 0| II or I! IV or. In. I! It It. It I! 00 I! I. 0.0. no .pto u b to o).or II II OI II In on I. to It on I. 00 0t 0' I. it II II It, OCT! II,‘I
%.lt m a . .
.w m g T
‘ . t
\J O 1 .1 X m \l O 5) 0 \l O 7 \I 0
mWme WMH mum?) 0 ie a m 2 06m 0M3 004 0
. 010b< 111. 11(. 1 mfimwn L c 11( 11( 11—.2( 1
L s a m .m 9 .
a tt mm 03
I. I. II II II 'I I. II II II L. I. I. II II I. II I. I‘ I. I. I. O. a s I. O, .4 II II II I. II II I. I' II I. I. II I. I. I. 'I I. II II III .I I.
Shi n .
n. . e td .1
H-“ Sm 603 max t. .
_ . . . .m b 0 .1 s
S 9 t
H Er» f
.m n C 0 .m
m e150 1 a
.1 s MMfi 1 L 9 9 We
L on Ow 0.» we a o e .m S 9. 4. 9 - 9 .
s W 9 9 9 a T c s 1 l v . r
s 1 7 .5 3 r Me v. a EU _ 9 4 e
a _ . . . e h l 1 O . . v
1 O O M m v t a C 1 l 5 0 A
C 8 6 A f e .
0. on. on_ to to no on I: o- 0. tv o. I. 0' 19 I. on In It I. O. l. I. II 9. 0|
0| .- I0 I. '0 II to no n. it 0- no. 00‘ at 0- no .00 II It on o. I. O. 09 n
. e d
damn 626.53.. mo mmMpmmohwm t a m EBB 80.; 8.3939

 

”223

Tablesl, J and K put in parallel columns the results of‘the.group-’”"

ing on one variable of each of the variables with land values and of the.
two cross-tabulation correlations for each. .

 

 

 

 

 

 

 

 

 

Table r
Percentage : Average : Ratios of Values :
of Improved : Percent : Grouping on : Distance { Soil Type
gand : Improved : One Variable : Constant : Constant
20439.9 : 29 : 100 : -lOO : 100
40.59.9 : 52 : 222- z "240 : 202
50.79.9 : 66 : 336 : ~286 : - 263
FO.99.9 : 88 : 420 z 295 : 275
~ Table J 1
Distance : Average Dis- : Grouping on :Pcrcentage of: Soil.
from Town ' : tense from- : One Variable :Improved Land: Type
' : Town : ‘ :COnstant : Constant~
0-4.9 : 2.2 . 100 : 100 : 100
5'909 : 7.3 = 76 3 85 3 88
10-14.9 : 12.4 : 65 : 76 : 69
_;5-13.9 : 16.9 : 44 : 55 : 42
Table K .
Soil : Grouping on : Distance : Percentage of
Type : One Variable 1 from Town : Improved Land
: ' : Constant : Constant
Loam V: 1001 : 100, : - 100
Clay : 72 z 81 : 82
_§§ndy : 64’ : 78 : 68

 

The next step required is to hold two Of the independent variables
a constant simultaneously and see what effect is left for the third. Since

; all three variables are correlated, the ratios holding only one constant
all show too pronounced effects. This could be done by splitting each of-
the squares in Table E into three parts for the three soil types and summar~.
izing the results as before; But three times as many farms would .be needed
to give even as stable results as with two -independent variables at a time.
If a four-th variable were involved in the problem, the squares could be

. split again} but this would-again multiply the number of cases needed. This
points to a mjaor limitation of the cross-tabulation method of correlation.
Many observations are needed if more than two variables are to be isolated,
especially if the results are to be stable.

229

Another possible_method of attack on this problem is that of
partial association outlined in the preceding section. It will give
definite coefficients.of_partia1 association but no rates of regression

as ordinarily presented.

A third method of attack isuthat of linear or curvilinear mul-
tiple and partial correlation. The linear treatment provides no way of
handling satisfactorily,the qualitative factors such as soil types, unless
enough cases are available so that a separate analysis can be made for each
soil type. A procedure sometimes employed is to assign arbitrary measures
to the qualitative classes, preferably in the order of their effects, ap-
proximating their'true effects as nearly as can be surmised in advance.

Let us suppose that the_following measures were asaigned;p,Sandy 1: Clay 2;
Loam 3. a The multiple correlation-coefficient'obtained would be raised if
using these rankings did not introducegerrors‘of measurement_greater than
the effect of soil type itself.3 These would.be the'further errors arising
from putting individual farms'in the wrong Soil classl' If the errors of
measurement were large, the not correlation.coefficient‘forlsoil type would
.also be neglible. ‘So far‘as the_regression equation is concerned, the re»

sults would reflect the usual effects of errors in an independent variable*.

_ ‘When curvilinear multiple correlation is used, the effect of
qualitative independent factory upon the soil type can be measured fairly
accurately if enough observations are available. The different soil types
for example, would be ranked according to theremexpected effect; Then,

when curves were fitted by the'apprbximation method, if the excess of

a

‘ .loam over clay was greater than the excess of clay over sand, that would .

.Lput a bondnor lump in the curve_at the point corresponding to clay. If
:there were‘seVeral-different cagegories, the curve.would become only a
:brokon line, showing.the separate relation of each qualitative category
upon the independent variable, . : ‘vr%'<‘ ' ' '

The data of the 80 farms were anhlyzed by multiple methods, soil
type being included as one variable with the following scale: loam 100;
clay 82; sandy 73, This was obtained by averaging the two series of ratios
in Table K with location and land improvement respectively held constant.
It was realized that these measures exaggerate the effect of soil type, but
no better measures werefavailable."""' ’

The results obtained were as follows!**'-'

Xd=Land value per acre, .‘ .'0
X1=Percentage of improyed land'~ . .
XgaDistance from toWn-- ”"- -'
K3-Soils ~‘Clay 82: Sandy 72; Loam 100

Gross r - XOX1-.82 ‘_ , net r: xox1;xéx3_4 €96
Xena-.77 . ; ' -: xoxa.nx3-’i .79.
‘ X°X3-.46 . H a , ..x°x5;x1xga'**,791

. 2 ' ~ ' (a .v ;' . ,

 

‘.9: The lubJoot or the effect.ot error. of measurement 9

V -w sults is described'hy Dr. lfieknel-inraalhter'sectionv ., , .
a ** Multiple correlation method is discuheed in later sections. .

n correlation re—

230

Net regression coefficients: XOX1.X2X3- 1.774
‘ xoxz.xlxg- 3.809
XOX3.X1X2- 1.538
Multiple correlation coefficient: 3 a .9688
Regression equation:
Xo- 1.774X1- 3.809X2- 1.538X3a 59.09

The net'correlation coefficients are all higher than the corres-
ponding gross coefficients. It is interesting to note how much higher the
net coefficient is than the gross for soil type. Apparently in the simple
cerrelation analysis the other two variables largely covered up the effect
of soil type. The multiple coefficient is very high - higher than would
probably result with actual data. Table L attempts to bring together for
comparison the rates of regression resulting from the analysis thus far.

The range in percentages in Tables I and J have been reduced to averages,
using the average of the last two groups as the upper limit, and then con-
verted to dollars per acre. The regression rates have-been greatly reduced
from the simple correlation stage, and also from the cross-tabulation stage.
Soil type has absorbed some of the difference, but more of it has disappeared
because no longer assigned to more than one variable at a time.

Table L.-Bates of regression.
Cross-Tabulation :Multiple

 

Simple Holding One Variable :Correla~

Correlation Constant at A Time :tion
Percentage of land improved 2.46 2.0 2.2 1.7?
Distance from town — miles 7.46 4.7 6.3 3.81

. Soil type-scale of percents

-. .. on e. - 43 a. a. I. u a. o,
a. an n o. no on n. o. a- an - o-
n. .. no .. u ‘- a. no

.. n- I. I. .0 ea .. no u

 

The next step is to apply the regression equation to each of the
80 farms and determine the residuals. Twenty-six residuals were less than
$5, 25 between $5 and $10, 15 between $10 and $15, 7 between $15 and $20, and
7 were over $20. This was possible with a multiple correlation as high as
.9688.

Comparison of_these residuals with the original data shews that
the six farms in one group along a lake shore average plus residuals of $15
per acre. The four farms within two miles from town are perhaps forty
dollars per acre higher than those nearest them. = It is proper to ask the
question whether-these farms should not be.dropped-from the analysis since
there are not enough of them in the sample to make a separate variable for
them. The residuals for them are significant, but not accurate determina»

tions since not derived simultaneously with the other three variables.

231

~ Ayeraging the residuals by soil types indicates that the scale V
used places the sandy and loam soils each a little high and the clay soils
too low. The following scale-would have fitted better: 100,89,69; that is,
if a straight line is to be used to describe the regression. It must be
remembered that these measures for soil types are merely relative to each
other, and that the regression line is fundamentally only a function.of the
scale of values assigned to them, although partly substantiated by the
cross-tabulation analysis. If distance from town and percentage of land
improved could have been held constant simultaneously. the scale would have
been, let us say, something like 100,92,80: and the regression line would
have been steeper in proportion.

It will be remembered that the cross-tabulation analysis stronge
1y suggested non—linearity in the effect of improved land. Plotting the
residuals for this variable and averaging them by sections along the re-
grossion line reveals a small amount of this still. At from 50 to 70 per
cent of land improved, the land values average about $5 per acre above the
regression line, and above 80 per cent, about $8 under the regression line.
Most of the apparent curvilinearity of the cross—tabulation analysis proved
to be due to soil type and location_thnt were correlated~with percentagp
of land improved. One cannot be sure that the curvilinearity still remain—
ing is not due to other variables not included in the analysis that may be
correlated with land improvement, or that do not average out at all portions
of the regression line-because of the fewncss of the cases{

The residuals suggest a downward inclination of the regression
for distance from town beyond 15 miles. One strongly suspects that this is
due to the defect in the scale for soil types mentioned above.

It would seem advisable as a final step to repeat the multiple
correlation analysis with the corrected scale for soil-types, and omitting
some of the abnormal cases. .If_the residuals still indicated curvilinearity
it might be desirable to employ curvilinear methods.“l '

' A method sometimes used for handling qualitative factore.is #0
leave them out of the first analysis and correlate the residuals with them
by grouping them by quality classes. This would not have given dependable
results in this case because soil type is intercorrelated with the other
variables. The multiple correlation coefficient with soil type omitted
was .88, and the regression rates were all too high.

In many analyses, the independent variables, or part of
them, are very little if any correlated with‘each other; and the fact of
‘this can be demonstrated in advance or at Various stages in the analysis.
When such is the case, tabular methods may-be adequate for all practical
purposes. In any case, most problems should be pushed as far as possible
.by tabular methods before a multiple correlation analysis is undertaken.
This is especially true with unexperienced analysts.

a;

 

* See later discussion by Dr. Ezekiel.

232

Elementary I It will not be necessary to discuss inlmuch~
time-series detail the use of elementary methods in time series
analysis analysis. The foregoing discussion is all pertinent.

Given two independent variables entirely unrelated to
each other and an analysis such as that of L. H. Bean's in Figure III will

serve.* Of course one would want many more than eight observations before—”

predicting with much confidence. Establishing that the two independent
variables are unrelated will not ordinarily by easy. and will usually be
impossible with only a few observations. If, for example, one wished to
determine the effect of both supply and industrial activity on the price
of cotton, one would probably have to reckon on the fact that the supply
of cotton and industrial activity are in some measure-mutually interre—

lated, and if so a good many years' observations would be needed to isolate a

the effects of the two. Such methods as Figure III represents would give
different results even then according to which was taken as independent
variable first and which was correlated with the'residualsg Let one as-
'sume that the price-of feeds is the major factor affecting supply of milk
and correlate the two and-assign the residuals to the price of dairy pro-
ducts and one will get one result; assume the opposite and one will get

a different.result. Even if there is no relation between them. one can
easily bank too much on the exact validity of the curves. A differently
placed supply price curve (1) in Figure III would have given a_different
trend curve (2). The two are obviously drawn in such a way that they
give the smoothest fit for both. It is very unusual even in normal times
to get so smooth a line of growth as curve (2) represents. One would
expect such things as the fluctuations of industrial activity to disturb
it. One surely would not say from the evidences of these curves that it
has not done so. ‘ ‘ 4 ‘

‘ Nevertheless such methods should be the first to be used in any
tine-series analysis. Given observations enough, cross-tabulation can be
used. Ordinarily, however. the cases are so few that one has to utilize
all of them as fully as possible in_order to get usable results, which
leads to the use of multiple correlation methods. raising extremely diffi-
cult questions of interpretations of results. which will be discussed later.

Variables to . , I CorrelatiOn can measure only the extent and
be Included** the kind of relations existing between the two or

more series of variables which are subjected to the
statistical study. The meaning of the results de-

. - - . . pends entirely upon the pertinence of the variables
selected to some particular problem under study.‘ Whether the variables
selected are pertinent or not is a matter of logic, and not of statistics.
Since. however, selection of the proper variables for study is the first

~ a critical point in the study, a case will be given to illustrate some of

L1the logical‘processes inVolved‘in-such-solectién.

Suppose the proprietor of a cafeteria wishes to determine why
the number of patrons varies from day to day. Let us further suppose that
the lunchroom‘is in a business district. and serves only mid-day meals.

 

-- 'Chart, III from a paper read by Mr. Bean before the New York Chapter
of the American Statistical Association, April 20, 1938.
** By Dr. Mordecai Ezekiel, Bureau of Agricultural Economics.

233

Then the dependent variable — the one the variations in which are to be
explained — is the number of lunches sold each day. This is shown to the
left of Figure l, which figure may be called a "factorgraph".

The factors which might cause the number of patrons to vary may
next be separated intwo two groups - those regularly recurring and those
irregularly occuring. The first group will obviously include time fac—
tors — these which cause regular increases or decreases in number of
patrons from day to day, week to week, or seasonally through the year.
we may go back of these, however, and try to find what causes them. A
half—holiday on Saturday, for example, may explain low sales that par-
ticular day of the week; decreased patronage in the summer may be due to
hot weather or to absence of patrons on vacations; an upward trend in the
number of patrons may reflect a steady growth in the number of persons
within reach of the store, decreasing competition due to competitors go—
ing out of business, or a progressive business which is out—running its
competitors.

Holidays, such as Easter, July 4, etc., which occur at stated
periods, may also affect the business for better or worse, and must be
considered.

The irregular group of factors, however, offers the greatest
variety. Factors that may affect trade irregularly may be grouped under
several heads; (a) those affecting the number of potential customers with-
in reach of the lunchroom; (b) those affecting their economic status; (0)
those affecting their desire to eat lunch indoors on any particular day;
and (d) those affecting their decision as to whether to eat in this par-
ticular store. Each one of these heads may be further subdivided into
individual items, as illustrated in Figure 2.

After laying out all the related elements in the problem, and
carrying them out as far as possible, the next step is to select variables
which will serve as numerical representations of each item. After these
values are secured the relationships among the several groups can be
studied, using the factorgraph as a controlling guide as to the particular
influence each individual variable is selected to measure, and the way in
which its influence is expected to be exerted. Then when the statistical
analysis of the relations among the various factors is completed, the in—
vestigator will be able to interpret logically the significance of each of
the variables whose relations he has measured.

Statement Once the variables to be employed are selected,
of Variables* the next problem is in what units to state them. In

studying land values, for example, the value of a

given farm may be stated as total value, as value per

acre of all land, or as value per acre of improved
land. Which one to select depends on what other variables to include and
how they are to be stated. On the per—farm basis, the total value of the
farm might be correlated with the value of the dwelling, the value of other

 

* By Mordecai Ezekiel.

.dwhwpwwuo a mo mmqumnp mqwuommmw whovomm now 3 nmmhmpouodm .H ondmwh

when and mhmnkwmam cayenne“ mooH>umm /
umHnnpm

owoum on» nH was chuckmwam campanu mmoHpm

macaw nunsa mafipumaoo mo aofiuwooa can nmpasz

”noun Haas
Iothmm on» on ma nowmwomn.

oqwnmnfiw

Hmequm . mnwnquH on ma mmpwmon
mhdpwnomeoe
mmH>HH mo pmoo aw . mmeOHmSo haanmuan

and Ho>oH chnm awapeh.qH mmwndno HdeQopom Mo mspaum ofiaonoofi
dm>wmomu wows: aw mawqdno

/—.——_.~.

. ‘ haw new
manmonoo Boa mo pqmanmwapdpmfl mhmao»wso HaHpnopom mo dunks ‘ . . // cophom wonoaun
mmufioflfidp kwn mo nowpomum , . ‘ . _

comOHmEo muonssn aw omnaao
meOpm mdmpomaou aw ducks , . unmanau
. camps.

mkofiovmfio Hawpnopoa mo.umpEdu nw cache

.Opo «mmac hum .mhuuHHoA HwHooam

maowpuodb no unoppam Mo wonomnd

anomdm . x
w w mafiphaoo»

a assume: dw womnwnozamnomuom _
hflpaagwmm

#mmkloplxm m3

.. . finSHom W mm? 3..th

235”

buildings, the acres in cultivated land, the acres in pasture, etc. This
would tend to show the contribution pix; acre ofeach. of the acreage ele...
ments, and should tend to give a high correlation, since under normal?

conditions the value of the farm may be expected to approximate the value

of the buildings plus that of the several tracts of land.

But if it were desired to measure the influence of_type of
road, fertility of land and distance from tawn, on land value, they could
not be so readily included in the same additive equation. To do so would
assume that a given increase in fertility or yield per acre would add the
same amount to the value of the farm no matter how large or how small the
farm was. ' ‘

_ If the value were stated as value per acre, that would partly
solve the difficulty, for a given change in fertility, distance from
town, or type of road, would then be assumed to have the same influence
upon value per acre no matter how large or how small the farm use. But
that would introduce difficulty with other variables. The dwelling, for
example, would not become larger in direct proportion with the size of
the farm. Very large farms with good dwellings would have a very low
"value-of-dwellingsuper-acre," and small farms with poor dwellings would
also have a low "value—of—dwellings-per—acre." Only some method of
determining the effect of value of dwellings on land values separately
for farms of different Sizes would take care of this difficulty.*

The case mentioned also illustrates the need of something
other than an arithmetic regression equation to express certain cases.
If it be assumed that the_more fertile the farm. the greater the effect
of nearness to town would be, and that the nearer to town, the greater
the effect of an increase in fertility would be, that could not be ad
equately expressed by the regression equation: '

Value per care - f(distance) f(fertility) etc.

Instead, the multiplying effects of the two variables upon the Value could
be allowed_fgp by using the equation:

Value per acre = [jf1(distancei] [f2(fertility2’l [f(etc.i) '
which, for the actual process bf computation, can be stated:

logarithm (value per acre) = ¢1(log distance).f¢3(log fertility)f

fl3(log etc.)

This logarithmic equation, which puts the relations on a
relative or proportional rather than an absolute or arithmetic base, is
',a very flexible one and One that can be used in a great many types of ,, n

problems. ‘

 

* See the appendix of U. S. Dept. Agr. Bul, 1400, Factors Affecting
Farmers Earnings in Southwest Pennsylvania, for an example of sta».
tistical treatment of a problem of this type. .

236

One further consideration with respect to the statement of
variables which needs to be mentioned is the danger of false results or
spurious correlation if the variables are improperly stated. Thus, for
example, if an'attempt were made to correlate the value of farms with
three factors,~namely, (a) the percentage of land in corn, (b) the per-
centage of land in wheat, and (c) the percentage of land in all other
uses, it would be impossible to solve the problem, or else it would
give a spurious result, because the three factors (a), (b), and (0)
would add to exactly 100 percentage in each individual case, and after
variations in (a) and (b) had been held constant by statistical means,
there would not be any roam left for variation in (0). Only by dropping
out one of the factors, say (c).would significant results be secured. The,
regressions on (a) and (b) would then also show the effects of (c); for
example, the increase in.vslue for each unit increase in (a) would mean
the increase due to substituting one unit of (a) for one unit of (c);'
changing the sign would give the effect of substituting one unit of (c)
for one of (a). The same principle would then apply as between (b)

and (o).
Choice of . -.V 'In using correlation analysis, there are two
Method* _ types of alternatives to be considered: first, as

to the number of variables to be used simultaneously;
and second, as to the type of relation betieen the
_ variables to be assumed, i,e., whether linear (a con-
stant unit change in the dependent variable for each'given unit change in‘
the independent), or non-linear. , H

Where the variations in the dependent variable are to be related
to variations in two or more independent variables, the problem arises as
to whether simultaneous (or multiple correlation) method must be used.to
determine the effect of each one. ' H

The simplest method of treatment is to ignore the fact that the
result is due to two or more causes acting simultaneously, and determine
by simple correlation the changes in the dependent variable for each change
in one casual factor, the the change for each change of the next causal
factor, and so on till one has been taken into account.

For this method to give an accurate measurement of the relations
which really prevail, two conditions have to be fulfilled; first, no cor-
relation between the several independent factors; and second, a large
enough number of observations so that the effect of the random.variations
in the other independent factors can be syeraged out.

This may be illustrated by a definite case. Suppose that the
yield of corn is being related to rainfall in July and to temperature
in July. Suppose further that there is no correlation between rainfall
not: and temperature. Then when a scatterndiagrsm between rainfall and yield.

,. . a ._._.._._l M “.“ggggi -.—,—-4-_--———» 3...;
jj:1_!3r,ubrdecei Ezekiel _.. ~

I .. . 4'»

.~.AA!

237

£33. cw opgmpmmsmp 95 “:2”. am 3“,?me nwmrump nowadamh 3;
was flgoo mo Shot” 98 ~33. 3 :mwnmmh 5253 H8332. Amy mflsoflm magma? pmpumow .m .mfm

m mwmpga. 95.5

 

 

 

 

 

 
 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

mask/Z” 2H GHAEZHflH mHmoZH 2H AQJEZHE
. N . 0 ¢ m N ‘ o
r m . E _ u .3
_ n .. m . _
. m. . m a w .
. _ m _ m _
.m w . . _ .
.. . w .
. allHldT IIWII . a u I QM u— m u o 0 I1 on
_, . m . . _ .
m m . _ . .
_ O . a . .
.u.|¢ ., . o. . -rti¢.1: _ _
{I'll . - . _ 1:- I1. om.
.. . . m _ Mm \ _ M \
a ._ _ o . _, A D “.7. 1 .
. . . \.\ _ u _
m .. . _
IIVlIllvb I IEMIIIIII I C I'll ll. om ~ Qt ‘ — m 0*
. _ . __. .W . 4 m
W W _ . . m m
M _ . “ ~ _ ~
. . A _ . w _
a. m . _ . M _ 804 3m
m . muonmmn .Q. . mamflmdm

.. 83¢anan 300 Mo Show».

238

is constructed (Figure 2,a), the differences in the average yield of
corn between the different groups according to rainfall are not in-
fluenced by temperature at all. This is because, with no correlation
between temperature and rainfall, the average temperature is Just about
the same from group to gmepp according to rainfall (Figure 2,b). Any
effects of temperature differences on yield are averaged out when the
yields are classified according to differences in rainfall. The regres«
sion line in (3) therefore shows the changes in yield with changes in
rainfall, with temperature remaining at the average, Similarly, a
regression line for these data showing the change in yield with change
in temperature would have the effect of rainfall removed; it would show
the change in yield accompanying change in temperature with rainfall
remaining at the average.

On the other hand, with data where there is Significant corre—
lation between independent variables, it rs necessary to use some method
which determines the change in the dependent variable for changes in each
one of the independent variables while simultaneously eliminating the
effects of the other indepdent variable. The method of multiple correla~
tion comes the nearest to doing this satisfactorily. Whether it can do
it completely satisfactorily or not (in addition to the question of line-
arity or curvilinearity, which will be taken up separately) depends upon
how well the form of regression equation used really expresses the way in
which the influences of the several independent factors combine to produce
the variations in the dependent factor or result. If the effects of the
several independent factors simply add together to produce the result, then
the simple additive regression equation:

x - a+ f2 (x2) + f3(X3) + (etc.)

can‘Pe used to determine the separate effect of each factor. If instead
the effects of the several factors multiplying together to produce the
result, then the logarithm equation:

log X1 = a +-f2 (log X2) +gf2 (log X3)‘T (etc.)

will most satisfactorily determine the relations. If, finally, the effects
are such that it in the combination of the several independent factors which
produces the result, so that there is a particular unique effect upon the
dependent for each particular combination of the independent factors, then
still more complex equations of the form:

X1 = f(x2, X3, X4) etc.

are needed to determine the way in which the dependent factor varies with
the different values of the several independent factors.‘

 

* A fuller discussion of this last case, and methods of measuring such re-
lationships, are given in "The Determination of Curvilinear Regression
*Surfaces' in the Presence of other Variables," Jour. Amer. Stat. Assoc.,
Sept., 1926, pp. 310-320; and a non—technical discussion of their character-
istics, in "Factors Related to Lamb Prices", Jeur. of Pol. Econ., April,
1927, pp. 236-239; both by Mordecai Ezekiel.

239

As regards the question of whether linear or curvilinear methods
are to be employed: linear methods are applicable only when it is reason-
able to assume that a given change in the independent variable will cause
the same corresponding change in the dependent variable no matter how far
it is carried. While relatively infrequent in agricultural economic'studies,
there are a few cases which are truly of this sort. Thus the area of peas.
ture needed is apt to continue to increase with the number of cows to hm
pastured (fertility being held constant): the price of hogs at country

\‘.points is likely to continue to change in the same relation to the prices

at central markets; and the price of hog products at wholesale is likely
to Continue to change with the price of hogs at wholesale; all no matter
how far the changes are carried. But in most economic or agronomic rela-
tions the reverse is true. Production does not increase indefinitely
with additional increments of fertilizer, at least certainly not in the
same proportion; profits do not increase indefinitely with larger fanmn
or greater capital: yields do not increase indefinitely with heawier refine
fall or higher temperatures; nor do prices continue to fall to the same .
extent with successively larger and larger production. In such cases the
assumption of linearity might completely distort the actual relations, or:
even result in a conclusion of "no relationship" in the case of such re--
lotions as that shown in Fig. 4.

When it is decided to assume curvilinear relations, the next
question is whether to determine them by the free-hand graphic method or
the use of a definite equation of some sort. The result obtained by the
free—hand curve is an empirical description of the relationship - but so
is the result obtained by a definite equation. unless the equation used
convoys some lo cal ex ression of the theoretical relation of the var a
ables involved. The way in which an equation may express logically some
characteristics of the relation has already been discussed under the head
of "curve—fitting". UhleSs the constants of the equation when determined
do have some logical meaning, it is worse than useless to use a definite
equation, for the function as determined by the equation may distort the
true relation as badly as the straight line would in Figure 4.

One further caution as to choice of method needs to be stated:
none of these methods provides reliable or stable results unless there are
enough observations. When curvilinear methods are used. the significance
of the relations varies at different portions of the curve. The results
may be entirely dependable within the central portions where the bulk of
the observations are concentrated, yet have but little reliability at the
extremities where there are only a few cases to indicate the shape or posi-
tion of the curve. Inkewise with multiple correlation: it takes more ob—
servations to determine the relations accurately when the result is due to
a single cause. Further, the more closely the several independent variables
. are inter-related, the more observations it takes to measure the effects

'fof each one Separately. Particularly in multiple curvilinear correlation.
if the several independent variables are highly inter-correlated it becomes
very difficult to determine the exact influences to ascribe to each one un»
less a large number of observations are available. With insufficient ob-
servations in such eases, the major affect may be ascribed to whichever
variable the investigator pleases, since if it is accounted for by one, it
is then removed and will not be left to be credited to the other. If two

240

variables are perfectly correlated it is impossible to determine what
part of thezesult is due to one and what part to the other. If they are
closely correlated, only a large number of observations will make separa—
tion possible. In time series problems several elements are often
trending upwards or downwards simultaneously, and with the limited number
of observations usually available. there is no absolute method of de-
termining what part of the total effect is due to one and what part is
due to another. .

No absolute rules can be laid down for the working limits of
reliability as to numbers of cases. The usual measures of probable or
standard error, supplemented by nStudent's" and R. A. Fisher's tables of
probabilities for small samples, Smith's formula for taking account of
the influence of the number of variables on the apparent closeness of
correlation, and Yule's formula for the standard error of net regression
coefficients, should add in particular cases.*

Where multiple curvilinear correlation is employed using free-
hand curves, no definite criteria of reliability have yet been determined.
It is evident that these results can be no more reliable than similar re-
sults with linear relations, and in most cases are probably less so - that
is, it would take a larger number of observations to secure equally valid
results using the same number of variables. As a rule-of—thumb guide,
Smith's correction of the multiple correlation coefficient for number of
variables might be employed, reckoning each variable whose effect was
curvilinearhy measured as equal to two or more variables, depending rough-
ly on how meny constants would probably be required in an equation to
erpress'the shape of each of the functions analytically.

Thus if with 4 independent variables, only 80 cases were avail-
able, and a multiple correlation of g a .80 had been obtained, and it were
decided that an equation of three constants (such as a cubic parabola)
would be required to express each function, then the 4 functions would be
, equivalent to 12 linear variables.

 

‘ Fisher, R. A., Statistical Methods for Research workers.

Smith, 3. B., Forecasting the Price of Cotton, Jour. Amer. Stat.
Association.

Yule, G. U.. Introduction to Statistical Methods.

241

Applying Smith's formula:

13'

 

Where m - no. of variables
n a no. of cases

p a multiple correlation observed

WI
1

j probable true multiple correlation

In this case the equation gives:

ml‘.

 

0.67

The adjusted correlation, 0.67 indicates less than two—thirds
as close a relationship as the apparent correlation of 0.80, since the
closeness of relationship is shOWn by the squares of the correlation.co- ;
efficients, which are 0.40 and 0.64, respectively,: - '

Purely as a rough rule—ofethumb, it may.be said that simple
correlation results (2 variable) cannot be expected to have much stability
unless there are at least 20 observations to work with; linear multiple
correlation, with 3 or 4 variables, unless there are 30 or 40; and multiple
curvilinear, unless there are 40 to 60. Where absolute limitation of the
universe makes it necessary to work with smaller samples, then the rules
of error previously mentioned should be referred to in interpreting the,
results, and the investigator should be duly modest in generalizing from
the results of his findings. . f ' ~ “ '

S'gnificance of H Correlation results may be used either as a

Results as Descri ~ description of the relations among variableS‘
tions of Date* . , in the particular data used, or as a basis for

, generalization as to relations in the universe
lfrom which the date were drawn. Generaliza»
tion involves all of the problems which are treated statistically under
the general head of "sampling". This present section will'discuss the
meaning of the several-different ways of expressing correlation results
solely as descriptions of the date studied, leaving aside the larger
questions involved in generalizing from them to other data.

* By Mordecai Ezekiel

242

'The several different results of correlation analysis may be
named as follows: The symbols used for the several different constants
are those used in the texts of Yule and Mills, and in the teohnical
papers of Tolley, B. B. Smith, and Ezekiel.

§ymbol Name

Simple linear correlation:

 

 

 

bxy __---- ________________ Regression coefficient of X on Y
rxy -- Correlation coefficient of X with Y
X I a +‘bxy Y ———————————————— Regression eguation of X on Y
sexy orag‘ ---------------- Standard error of estimate, in estimating

 

X linearly from Y.

Simple curvilinear correlation:

 

 

X 8 i (Y) ---------------- Regression curve of X on Y
ny -- Index of correlation of X with Y
Se OI‘a”z ----------------- Standard error of estimates, in estimating

 

X curvilinearly from Y

Multiple linear correlation:

b12.34..n -------------- Net regression coefficient of X1 on X2, variables

X3 to Xn held constant

R1.234..n ————————————— Multiple correlation coefficient of X1 with x2 .
to Xn, together

D1.234 ----------------- Determination coefficient of X1 by X2, X3, X4,
etc. combined.
r12.34,.n -------------- Partial correlation coefficient of
X1 with X2. variables Xx to Xn held constant
d12.34..n ————————————— Partial detennination coefficient of X1 by

X2, the effect of X3 and X4 being allowed for.

243

X - a I b
1 M f 12.34..n 33'5"” 13.24.,n 3.3 +(eto.) Net regression cognition of
X1 0n Ks, X3, etc. to x11.

86" 01‘ka —--— ----------- Standard error of estimate, in estimating
1.234 ;, ——.-
‘ X1 linearly from X2, X3, X etc. to Kn.

X1 ‘ f2 (35) --4; ------------ Net regression curve of X1 on X2

- 1H.234: ..n -—-——-—g-—----—--Index 3: multiple correlationbf
X1 with X3, X3, etc to Xn.
X1 = a.¢,f3(X ) +-§3(X3) etc. Net regression equation of X1 with X2, X3,

etc., according to functional relations.

Se1 234 orZre---~—-e-~-~-- Standard error of estimate for estimating
0

X1, for X2, X2, etc. to Xn, using curvilinear

relations.

In spite of the large number of different symbols, due to the
four-way classification as simple or multiple, linear or curvilinear, all
of these different measures may be grouped into three different classes.
These are, first, the measures which show the change in the dependent
variable for given changes in the independent; second, the measures which
indicate what proportion of the observed variation in the dependent vari-
able can be accounted for by variation in the one or more variables which
have been related to it; and third the measures which indicate within what
limits of error the values of the dependent variable can be estimated when
accompanying values of the independent variables are used.

The measures of the first type may be illustrated by reference
to figures 3(a) and 4. Figure 3 shows, by a straight line, the average
yield of corn for any given inches of rainfall. .These yields may be
mathematically expressed by the eguation 2: a straight line. Y-- a byx X,
where Y gives the yield of corn, and X is the inches of rainfall. This
equation is known as the regression or estimating equation. The coefficient
_, byx'-which shows the ayerage increase in yield for each increase of one inch
‘. rainfall, is known as the regression coefficient of yield on rainfall. If
the analysis were turned around the other way, so as to estimate rainfall
knowing the yield, the regression coefficient would then be " showing the
average difference in rainfall with each increase of one bushel'in yield, and
would be termed the regression coefficient of rainfall on yield. It should
be noted that b xy ordinarily cannot be computed from byx. but must be de-
termined separately by a separate_computation. Only in cases.of perfect
correlation are the two exactly related; then one is the reciprocal of
the other.

2““!

 

 

 

 

Corn Yield Temperature ,
Bu.Per Acre A Degrees .... B
I ! f I
‘ I I
1 |
i I ,-
0 ' -'i
40 4

 

 

 

 

 

 

 

 

 

 

 

 

 

 

35 ' ,
o
. .l
o O I
O . l
30 .._.-__-iif.__
a g
i i
250 1 2 5 4 750‘ 'fi 2 3- 4
Rainfall in Inches Rainfall in Inches

c g
U.S.Dept.of Agriculture “rjdp Averages Bureau of Agricultural Econ.

Fig. 3. Scatter Diagrams showing (A) relation between rainfall in July and yield of
corn, and (B) relation between rainfall in July and termerature in July

Corn Yield
Bu.Per Acre

 

 

 

 

 

 

 

 

 

 

 

 

 

 

20

0 2 ' 3‘ . .,
Rainfall in July, in Inches

Fig. 4. Hypothetical change in yield of corn with differences in rainfall.

245

Where there is a curvilinear relation between the variables,
such as illustrated in Fig. 4, the changes in one variable for changes
in the other are shown by the regression curve, Y = f(X). This curve
may be shown either by an analytical function, such as Y = a b X cX ,
from which values of Y can be computed for any given value of X, or by a
graphic curve, or by a table from which values of X may be determined for
any particular value of Y. In Fig. 4 for example, the average yield of
corn is shown to be about 25 bushels for either 1 inch of rainfall or
6 inches, about 30 bushels for 2 inches, etc. Where there really is
such a vurvilinear relation, the regression curve obviously gives a more
accurage indication of difference in yield with differences in rainfall
than would the straight reggession line.

Where there are two or more independent variables involved,
the regression coefficients have a parallel meaning. For example, if
in the equation:

3‘1" 8‘ b12.3 x2 b18.2X3

X1 is yield of corn in bushels, X2. is rainfall in inches, and 13 is
temperature in degrees, the regression eguation as a whole would show

the average yield of corn for any observed combination of rainfall and
temperature. The constant b12.3, known as the net regression coefficient
of XI on x2, shows the average increase in yield for each one inch in-
crease in rainfall, the effect of changes in temperature being removed

or allowed for; and similarly the net regression coefficient b13 2 shows
the average increase in yield, X1, for each one degree increase in temper-
ature, X3, with the effects of differences in rainfall allowed for or
-taken into account.

Where these net relations have been determined as curves ins—
stead of straight lines, the estimating or regression equation is
stated: .

X1 = a £2 (X2) £3 (X3)

Here the value "£2.(Xg)" indicated the particular change in yield for
each particular difference in rainfall, the effect of temperature being
removed, and "£8 (X3)" indicates the particular change in yield for
each particular difference in temperature, the effect of rainfall being
held constant. Again the equation (x1) as a whole indicates the aver-
age yield observed with any particular combination of rainfall (X2) and
temperature (X3) according to their separate curvilinear effects, rather
than their linear effects.

The methods of estimating values of one variable from known
vvalues of one or more other variables, Just discussed, form the basis
for all the other measures of correlation which have been stated.

_ The second group of measures all indicate the closeness of
the relations by showing what proportion of the variation in the inde-
pendent variable can be explained by the one or more variables which

246

hare been related to it. This may also be stated as what proportion -
of the original variation can be reproduced in estimates made from the
other variables. These measures may be divided into two sub-groups;
(a), those which determine the proportion by stating the standard de-
viation of the estimated values as a percentage of the standard deviap
tion of the original, and (b), those which state thee sguare of the
standard deviation of the estimated variable as a percentage of the
sgggrs of the standard deviation of the original.

The first sub—group includes the coefficient of correlation,

r , which measures the proportion for straight-line relation between
two variables; the index of correlation , ih1oh measures it for
curVilinear relation between two variables, according to the particular
curve employed; R1 334 n the coefficient of multiple correlation,
."shich measures it for the relation between one variable X1, and the
two or more others X2, X3, etc., where only linear net regression lines
are employed, and P1 234 the index of multiple correlation, which
measures it between X1 andn several other variables, where regression
curves are employed to represent the net effect of one or more variables.
for both the coefficient and index of multiple correlation, the symbols
'to the right of the decimal point in the subscript indicate the Vari-,
, ables which are being used to estimate from, and the one subscript to the
left of the decimal point ind.icates the one variable whose Verne is be-
ing estimated. . . .

\\ , V _ ~
' There is one further coefficient of the firs t sub-group, the
coefficient of net correlation, Ti? 34 which must be explainqg'sopar-
ately.\ This measures the closeness of relation between the we vari—
ables X1 and X2 when each is first adjusted to eliminate the effect of
correlated changes in X3 and X4 It may also be stated as a measure
of the extent to which the variation remaining in X, after X3 and X4
have been accounted for may be further reduced by introducing X3 as a
factor in making the estimate. However, it is not a direct measure
of the proportion, but a rather involved mathematical function of it.

The second subgroup of measures are given the general name
of measures of determination. This name is used because in certain
extremely simple cases, where one variable is known to be composed of
several elements some of which are and some of which are not represented
in another variable, the determination coefficient measures the pro-
portion of all elements in the first variable which are present in the
second. Where the variables are known to be causally related, the
measure of determination may be used to indicate .the extent to which
one variable is determined by another by showing what per cent of
the variation in one can be explained by the other. This is par-
ticularly'uSeful, because the difference between the determination
esoffieibnt and 100% always gives the proportion which is not account-
ed for by the Variable or variables concerned - which is not true in
the case of correlation coefficients. a

The determination measures, by definition, are all simply
the squares of the respective correlation coefficients or indexes,
and ther09bre parallel them in all cases. ‘ The measure €12 is the

247

determination of X1 by X2, in simple linear or curvilinear correlation,
while the meamire D1.234 is the determination of X1 by X3, X3, etc.,
in multiple linear or curvilinear correlation.

The value 13d34 2 however, has a meaning distinct from
that of the coefficient of partial or not correlation. It is instead
the partial determination of X1 byZX, and shows the proportion of the
total (squared) variation in X1 which can be accounted for by varia—
tion X2, while simultaneously accounting for other parts of the varia—
tion by the net effect of X3, X4, etc. The sum of the partial deter—
minations by X2, X3, X4, etc., gives the total determination by all
independent factors; thus—

D1.234 - 12‘134 13%} Miles

However, these values are frequently difficult to explain, and especial
care must be taken te interpret them properly.Ill

The third group. of measures all indicate how large is the
variation remaining in the dependent variable after the effects of the
other variable or variables are allowed for, measured not in proportion
to anthing else, but in absolute units of the being estimated. Thus
in the relation shown in Figures 3 or 4, it would show the average of
the number of bushels of corn by which the several actual yields dif-
fered from the expected yields. This average, however. would be com—
puted by taking the standard deviations of these differences. The
standard error of estimate, as it is called, is therefore simply the
standard deviation of all of the"errors" or failure to account for the
variations in the dependent factor in the group of records covered by
the analysis. About two—thirds of the "errors", or differences be-
tween actual and estimated values, will therefore come within the
range shown by the standard error, and about one-third outside of it.
Where these errors are nonnally distributed, this measure also in-
dicates about how frequently errors of two, three, or more times the
standard error of estimate will occur.

The same name is used, the standard error of estimate, no
matter whether the estimate is made from one or more variables, or by
linear or curvilinear relations. In each case it is simply a direct
measure of the actual variation not accounted for by the particular
variables used and relations determined.

 

* Fbr details of interpreting partial determination coefficients,
see Factors Affecting the Price of Hogs. U. S. D. 1., Bul. 1440,
p. 35; Factors Related to Lamb Prices, Jour. Pol Econ.. April,
1927, pp. 258—254, both by Mordecai Ezekiel; and Adjusting Hog
Production to Market Demand, by F. F. Elliott. Ill. Agr. Expt.
Stat. Bulletin 293, pp. 5544557.

2’48

memsoaoom Hang H833 mo ~82de inaugufluwd mo pnm§n&3.w.b
.mmmpma zogflmmmoo mo mg.
a5. wEEmBqH ~maqufim gfiag .m .mfim

mmflodH 5 HHmmmwdm manonH GMAHmmqfldm . . mmflonH 5.” HHdmnwam
« . m z m H m

 

  
   

 

 

 

“Tirxklflul. om

1mm

 

 

 

 

 

 

M mm

 

 

 

 

mmwnmmm 23

 

 

 

 

 

 

 

 

 

 

 

0: +nm ammém

 

 

 

Zmoo .mo QHMHM

 

249

The partioular usefulneSs of each of the three different
grasps of correlation measures is illustrated in Figure 5, which shows
3 sets of simple correlation relationships, with various assumed re~
sults.

Here the regression coefficient is larger in A than B. In
the first case an additional inch of rain causes an average increase
0f 2.5 bushels in yield, as compared with an increase of 3.0 bushels
in B, the second case, But in case A, practically all of the varia~
tion in yield is apparently due to rainfall, as-shown by the high cor-
relation (.90) and the small size of the standard error of estimate
(2 bushels); while in case B, apparently factors other than rainfall
cause most of the differences in yield as indicated by the low correla~
tion (.30)'and the larger standard error of estimate (5 bushels.) In
terms of determination, apparently about 81% of the differences in
yield are related to differences in rainfall in the first case, and
only about 9% in the second.

In comparison with A and B, case 0 has much less variable
yields, ranging from only about 28 bushels to 32 bushels, compared
with a range of 24 to 35 in case A, and 22 to 40 in case B. Afipar-
ently only a small part (16%) of the variation in yields is associap
ted with rainfall differences, as is indicated by the low correlation
(.40). An increase of 1 inch in rainfall apparently causes only a
0.5 bushel increase in yield. Yet in spite of this low relation,
it is possible to estimate yields more accurately, given the rainfall,
in this case than in either of the other two, as shown by the standard
error of estimate, 1 bushel as compared to 2.0 bushels for A and 5.0
for B. This is because the original variation in yields is so slight
that even the small relation shown to rainfall is enough to make it
possible to forecast yields more accurately than in either of the other
cases.

These three cases illustrate the relative place of each of
the three types of correlation measures. Case 3 shows the greatest
change in yield for a given change in rainfall (the regression measure);
case A shows the highest proportion of differences in yields accounted
for by rainfall (the correlation or determination measure); while case 0
shows the greatest accuracy of estimate (the error of estimate measure).
To which of these measures to pay the most attention in a particular
investigation, depends upon the particular phase of the investigation
which is most important; the amount of change; the proportionate import-
ance; or the accuracy of estimate. All hare their place and none should
be entirely overlooked or ignored.

EEEQEE_2§ Errors of observation in the data for the
§££2£§_12 , several variables, such, for example, as might be
Dais: , _ . due to inaccuracies in estimating or reporting corn

yields or rainfall. has a varying effect upon the
‘ significance of the correlation'results. The exact
effect depends upon two conditions: first, whether or not the errors are.
random and unbiased (that is, falling purely by changs.equally above and

 

* By Mordecai Ezekiel

250

below the true values of the variables and uncorrelated with the
values of the Variables); and whether they occur in the dependent
factor, the independent factors, or both.

If (1) the errors are random and unbiased, if (2) they

occur in only the dependent factor, and (3) a large enough number

of observations are available to allow the errors to Canoel each
'other out, then their presence will lower the correlation coefficient
and increase the standard error of estimate, but will have only a
negligible effect upon the slope of the net regression line or the
shape of the net regression curve. In either simple or multiple
correlation in an ideal case, the values of the regression coefficients
would be exactly the same as if there had been no errors of observa~
ticn in the dependent variable,_but in practice there would rarely be
xsuch perfect cancelling out of the errors as this.

4 In case the random errors are in one or more independent
factors, with none in the dependent, they would have the effect of
reducing the correlation, increasing the standard error of estimate,
and in addition flattening out the regression line or curve for the
variable (or variables) in which the errors occured to a less marked
slope than it should have. Stated another way, the regression co-
efficients for the inaccurately observed independent variables would
show smaller values than they would have.had if the data had been
accurately recorded. ’In addition, in cases of multiple correlation
the net regressions for those independent variables which had no
errors of observation might be slightly altered, either up or down,
due to the failure to make adequate correctionlfor the effects of
the variables which were inaccurately observed. Such differences
in these regressions, however, would probably be minor compared with
. the reductions in the values of the regressions for the variables
which were themselves inaccurately recorded.

In cases where both the dependent variable and one or more
of the independent variables have errors of observation, the effect
on the.correlation result is a combination of-the consequences men—
tioned above. The closeness of correlation will be reduced, and
the standard error of estimate increased; the regressions on the
independent variables which are inaccurately observed will be reduced;
but the regressions on the independent variables accurately observed
will not be affected by the errors in the dependent variable, which,
given enough cases, will tend to cancel out,-but only the the failure
to allow properly for the effect of such of the il..pendent variables
as were inaccurately observed. . '

The way that'random errors might actually work-out in
practice may be illustrated by an actual example.l Over the 17 years

.from 1907 to 1923, the monthly price of lambs shows a very high corre-
' lation with the price of wool and the price of dressed lamb. Using
X3 for the price of wool in cents per pound, X3 for prices of dressed
‘lamb in cents per pound, and X1 for the price of live lambs in cents
per pound, multiple correlation giVes for the 204 observations:
31.23-‘991' and-3:13.144x2+.354x3

251

To test what effect random errors would have had on this
correlation, two dice were thrown 204 times, giving random values
from 2 to 18. These values were then added to the dependent vari-
able, and a similar set of values to one independent variable, to
see what effect this would have on the results. In the following
tabulation, the notation " X—r e" is used to designate the variables
to whose values those "random errors" had been added:

 

 

Independent : Dependent : Multiple : Regression
variables ; variables : correlation : equation
X12 and X 3 i X 1 i .991 ; .l44x2 .354 :3
x 2 and x z- ; x 1 c Q .821 Q .112x2 .414 x3
x 2 and x 3 (3% x 1 .953 .1esx2 .277 x3
X; andx3 e; X1 6 .804 :.152xz .306 x3

 

- These results illustrate the principles set forth above.
The introduction of random errors into the dependent variable (X1)
reduces the correlation, but does not greatly change the size of the
two regression coefficients. It would apnear, especially from the
amount of the reduction in the net regression X3, that the errors in
this case may not have been completely randomly distributed and un-
correlated with Xl, even though determined by throws of dice.

,But the second modification, where the error is introduced
into the independent variable X3 instead, is much more striking. The
correlation is not reduced so much as in the first case, and the re—
grcssion of X1 on X2 is changed only slightly from the original value—-
and increased, as it happens. The net regression of II on X3.+ 6,
however, is only 3/4 as large as was the net regression of XI on X3,
in spite of the fact that the error introduced was only enough to
raise the standard deviation of X3 from 6.14 to 6.64.

The final case, with errors introduced into both X1 and X3,
shows the lowest correlation of any, as would be expected. .The net
regression of Xli+ s on x2 is but little different from what the re~
grossion of 11 on X3 was, while the not regression of X1 +~ e on X3-+ 6,
though slightly larger than what the regression of X1 on X3‘+ e was, is
still quite significantly lower than the regression of X1 on 13. The
regression-equagieneig his last case, where X1.t e is the dependent,
is not greatly/ rom w it was in the preceding case with 11 as the
dependent, in spite of the fact that one of the independent variobles---
13 ~- had a significant random error of observation in its values both
times.

252

These cases illustrate the extent to which random errors may
confuse the true relations if they are allowed to creep into the ob—
servations. Just how great an effect upon the results such random
errors will have depends upon the magnitude of the errors, the original
variations in the variables, and the closeness of the inter-correlsp
tions. While equations can be derived to show how great a reduction
in correlation errors of a given magnitude will produce, they are of
little practical use in economic work, since it is usually difficult
enough to determine whether there are errors of observation or not,
much less to determine what magnitude they have.

In the problem given, the significant values determining the
effect of the errors are as follows:

5x1 = 3.96 . 6x1 + e a 4.74
Fits-36.14 sz +e=.;6.64

For errors in the dependent variable, the relation between
the true and the apparent correlation are indicated by the equation:

 

2 , a
R = 2 if 2 2 .-,
1 e.2.8..n. 1R 1.2.3,,n ( 1 p) . R 1.e.3.n 1
+' 2 .27 5-2
' 76':
x1

This gives what the new correlation would be if the errors
were truly random, so that the new regression equation came out as
V identical with the old. In the problem given,'this'gives an expected
value for R. of .827, as compared to the .821 actually obtained.

The practical significance of the principles which are stated
here is that if there is known to be a large but.random error in ob-
serving some variable, that variable may still be used as the dependent
variable in a correlation study without making the regressions or es-
timating equation very far wrong, if determined with a large number of
cases; but on the other hand, any use of that variable as an independent
variable will be certain to yield results which fell short of the actual
relations.

In cases where the errors are biased, they tend to make the
results of correlation analysis more or less in error, quite regardless
of the variables to which they apply. If the errors tend either to
magnify or minimize the differences which actually exist, they will
have a parallel'effect on the regression coefficients if they apply
to the dependent variables, and an inverse effect if they apply to an
independent variable. There are so many different types of bias,
however, that no more definite statement of.the effects can be laid
down. . -

Random errors have the same type of effect in the case of
curvilinear correlation that they do in linear correlation, since if
they are truly random they will tend to be balanced out along all the

363

portions of the regression curve alike if they are in the dependent
variable; or tend to confuse the relations along the curve if they
are in the independent variable, and so reduce the differences ob-
served.. . .

Biased errors, on the contrary, may happen to be concena.
trated along certain portions of the range, and hence have a much
more marked effect at one point than at another. While this might
seriously disturb the significance of the curve, it probably would
have an equally disastrous effect on the reliability of the straight
line, About the only real difference between linearity and curvilin-

earity with regard to errors is that random errors in the dependent

be necessary to secure equally valuable results for a curvilinear
regression. .

Where, with random errors, there are not enough cases
available to ”balance them out", the effect of the errors is to
throw a varying amount of error into the conclusions, the exact
amount of the error depending on how closely the errors approach
being cancelled out. The illustrative case, where with over 200
observations the regressions were still changed somewhat, probably
indicates what may be obtained by.a combination of slight departures
from true "randomness" in the errors with a sample not quite large
enough entirely to eliminate all of the resulting instability. This
may be nearer to what would usually happen in practice than the theor-,
etical complete elimination of the errors in the dependent variable.

Correlation as Certain peculiarities entering into
Descri tion of , analysis of geographic series call for special
Geographic Data.' consideration. This is particularly true when

correlation is used to describe certain attributes

of the series.‘. The validity of correlation
methods which such data depends upon the purposes for which the data were
collected. One purpose, for example, might be to study the factors af-
fecting the price of a commodity. Instead of building up the usual time
series, extending ever, let us say, a period of fifty.yean, it might be
possible to secure data at fifty or a hundred localities dispersed in
Space rather than in time. Such series would present a cross-section
view of those factors believed to be related to the price of a commodity
at a given time. The variations between these series could then be
compared and perhaps described by correlation methods.

Results of correlation analysis of geographic data may be of
slight value, however, because of certain difficulties frequently en~l

positions in space". But geographic observations are ordinarily not
separated by equal spatial differences and are usually not "successive".
In the same way that trends in time obscure the relationships being
studied, so many analogous "trends" "due to successive positions in

 

* By Dr. A. G. Black, University of Minnesota.

254 . -

Space" obscure the relationships found in geographic data.* This dif-
ficulty is the more serious in the latter case because it is practically
impossible to remove geographic trends; whereas a number of methods hays
been devised to remove trends from time series.

If it were possible to secure observations at known distances
from some origin, the usual methods of fitting trend lines would still
not be applicable because the aesumption that the magnitude of the ob-
servations will bear some functional relation to the distance from the
origin is not true. An observation taken two hundred miles north of
the origin will not be the same as one made two hundred miles south.
Differences in transportation facilities, soil, rainfall, temperature,
length of growing season, type of population and countless other factors
of a similar nature will be somewhat different for each of the infinite
number of points equally distant from the origin.

Trend in time series is removed to correct for those unmeas-
urable and undefinable forces which appear to exert their influences more
or less uniformly over a period. It is assumed that these indefinite
factors exert an influence that can be expressed by some function of time.
Except in the event that observations in space are made successively, it
cannot be assumed that any set of forces affects geographic data in such
a way that its influence can be stated as a function of dispersion in
space. , -

It may be objected that even if it were possible to eliminate
the "geographic trends", the observations taken at different points would
not be useful in economic analysis because the relationships existing be—
tween the variables would probably differ from place to place for reasons
other than mere dispersion in space. This objection applies equally well
to time series. There is no more reason for-believing that successive
observations in time are representative of homogeneous conditions than
are successive observations in space. The major objective of much time-
series analysis, moreover, is to isolate the effect of these very devia-
tions of conditions from the trend. In geographic analysis, the maJor
objective usually is to establish relationships between the varying geo-
graphic factors and some dependent variable. The geographic "trend"
may or may not be removed first. In the studies by Russian geographers
of the effect of distance from market on prices and production, the pro-
cedure has usuallyibeen to remove the "trend" by Setting up a series of
concentric zones within which prices and production are assumed to be
the same and then explain the variations from this. In the United
States, distance is usually included as an independent variable, Just
as time often is in time-series analysis.

- The attempts to analyze land values by means of correlation
analysis illustrate the nature of the problem of geographic analysis.

 

* "—---—--the problems arising from correlation due to successive
positions-in-space are exactly similar to those due to successive
oocurence in timo----—" "Student",Biometrika, Vol.X. p. 179.

2.6

In these studies, distance from local markets is usually included as
one of the variables; but location with respect to distant markets

and other points of_influence is usually not included. It is assumed
that the individual factors exert an influence proportionate to their
magnitude regardless of the location in these latter respects.*

Mr. H. A, Wallace has uublished the results of a soidy showing the
relationship of certain reasirnble characteristics to land values in
Iowa in which not even distance from local markets is considered,**
The average land value per acre (without buildings) for each ed: 11,.

as reported by the 1935 Federal Census of Agriculture was correlated
with county averages of corn yieli per acre, percentage of land in corn,
percentage of land in small grains and neroentuge of land not ploweble.
A coefficient of multiple ccrrelation of .917 was obtained. ihc study
published by Mr. G, C. Hens in 1932, in which he used correlation an-
alysis in an sttenyt to secure a formula useful in ascertaining the
value of a particular farm in an area, included distance from local
market as one var ile.*** The other dependent variables were cost
of buildings per acre depreciated to 1919, the date of the study, a
land classification index, and a productivity index. The dependent
variable was actual sale prices. The coeffiCient of correlation was
.81. Inasmuch as the sales were dispersed over a wkolc county, they
formed a geographic series. Recently the Minnesota Experiment Station
has cor? laced the average value per acre of land and buildings for the
counties of the state as reported by the Federal Census of Agriculture
for 1925 with the value of buildings per acre,-per cent of land im—
proved, a productivity index, per cent of land in woodland, an index

of roads, and a pasture index. The coefficient of multiple correla»
tion obtained is .97.‘ ‘

 

If sufficient care is exercised'in selecting the variables
which are logically,releted with the dependent variable, and in securing
measurements of them, the results of such Correlation analysis should
Constitute a worthwhile descriytion of the relationships involved, one
that'is pur.5culcriy useful because it attaches definite numerical values
to the influence of each variable, A complete correlation analysis,
including the partial coefficients, gives u more definite and complete
description of the series of data and their inter—relationships than
could be possible by ordinary descriptive methods. ‘t one must not
let the size of the multiple coefficient lead him into thinking that
these six factors constitute a complete description of the structure
of land values. To begin with, some very significant factors, such
as distance to lOCul market, location with respect to main highways,
size and quality of local market,size of farm,conmunity orga.iration,
balance out largely in the county averages. These and other similar
factors explain the difference between the coefficients of "70 to .80
usually obtained with individual ferns as the units of observation and
those of .90 to .98 usually obtained with counties as the units of ob—
servation. A coeffi*ie”* of ,“4 wears rmvghly that 44 yer cent of the ?'

 

 

* That is, "pro'ortionete" within the limits of correlation analysis
as expressed »y the coefficients in the regreSSion equation.

*‘ Journal of Land and Public Utility Economics, October, 1826.

*** Technical Bulletin 9, Minnesota Agricultural Experiment Station,1922.

 

(If.

256

variations must be otherwise related; a coefficient of .94 that only
12 percent must be otherwise related. If these figures are accepted
as generalizations of only limited experience, it would appear that
the individual farm factors balancing out in county averages repre—
sent roughly a third of the total relationships appearing in the data
for a given state. Secondly, the influence of the larger geographic
trends has not been measured. Thus all the county averages in Iowa
are higher than those in states further west, and lower in states to
the east, because of relative distance from consuming centers, sup-
plies of capital and interest rates, maturity of the regionr and
other similar factors taking on the nature of general geographic
trends. In the totality of relationships involved in the land value
problem, these factors figure in an important way also. An analysis
of state averages would throw Some light on the extent of such relay
tionships.

The problem of geographic trend may be further illustrated
by the geographic relationship between corn and big prices. Two price
series were constructed for the band of states extending from Maine to
California. The farm price of hogs per hundred weight constituted one
series, and the farm price of corn the second series. Thb coefficient
of correlation was .72. But surely this does not represent a simple
direct causeaand-effect relationship existing between the series. It
seems rather to show that both corn and hog prices are high or low at
certain localities due to distances from the principal areas of pro-
duction. In other words, the correlation reflects the influence of
geographic "trends". Prices of both hogs and corn are high in the
East and in the West and low at the centers of production.

The following hypothetical case will serve to illustrate
in extreme form how geographic relationships may enter into correlap
tion analysis. let us assume that two commodities, A and B, are
produced in different localities, but,that A is used in the produc-
tion of B. Let us further assume that the production period is short
and that there are few alternative uses and few substitutes for A.
Under such Conditions the changes in the price of the two goods will
be closely related. However, in the locality where A is produced,
its price will be relatively low, and the price of B relatively high,
because of tran3portation costs. The opposite will be true of the
prices in the locality where B is produced. ' If an analysis of the
prices of the two commodities was made, it is clear negative correla~
tion would appear, even though, if it were possible to set up two
series showing the prices of the goods at the same place over a period
of time, without doubt there would be a high positive correlation.
The correlation coefficient from the geographic analysis would be of
slight service in showing the direct relation of the prices of the
two commodities. It would, however, be a useful descriptive device
.showing how the prices of the two goods behaved over a wide area.

The high negative correlation would tell us that as we moved from
one point to another. the prices changed inversely, and that high
prices of one commodity was associated with low prices of the
other.

a,

I

257

An extreme case of variations in geographic conditions af-
fecting the analysis appears in the relation between percentage of
tenancy in the United States and the average value of farm land. It
is generally assumed that the rate of tenancy varies directly with
the land values. However. the correlation coefficient is only .07.
But if thirteen southern states are excluded from the series, the cor-
relation for the remaining states is .58. The southern states ex-
cluded have a high percentage of tenancy and a low average land value.
This is sufficient to offset the opposite relationship found in the
northern states. The north and South have an entirely different econ~
omic, social and agricultural history. In consequence, factors affect-
ing farm tenancy are quite different in the two regions, or more pre-
cisely, the factors affecting the general level of farm tenancy are
different. The percentage of tenancy in a given state is determined
by two sets of factors. those which set the general level and those
which determine the position with respect to the general level. When
state average land values and tenancy are Correlated, the series con-
sist of obserVations affected principally by factors other than land
values which influence the general level. These other conditions
differ greatly from state to state.

If land values and percentage of tenancy by counties within
a_single state are correlated, the same results appear, although con»
ditions within a given state are more nearly homegeneous than within
the entire country. Illinois in 1920 had an average tenancy rate of
42.7 per cent and an average land value of $187.60 per acre. Mississippi
had 66.1 per cent tenancy and an average land value of $43.40 per acre.
There is no indication in this Comparison that high land values are
associated with a high percentage of tenancy. However, when land values
and per cent of tenancy by counties in these states are correlated, the
coefficient of correlation is .64 for Mississippi and .83 for Illinois.

At least a part of the inconclusive result with the state
averages is due to definition. The term "tenancy" connotes an entirely
different type of agricultural organization in the two regions. Geo-V
graphic analysis can always expect to encounter this type of difficulty.
The wheat whose prices in different areas are correlated is far from
being the same sort of wheat.

The foregoing discusSion would point to the general conclu-
sion that the accuracy of correlation as a description of geographic
relationships depends upon the uniformity of the data. Geographic
dispersion generally results in heterogeneity because the individual
observations are largely resultant of different sets of forces. The
heterogeneity may be due either to a changing basis of definition or
to changes in those factors which bring about the basic levels of the
variables being studied. Some of these conditions can be used as
variables, or analyzed by cross—tabulation methods, as illustrated in
an earlier section. But very often they cannot; or a very considerable
preportion of them cannot. Some of them may take sufficiently the
form of a continuum so that they may be looked upon as "trends"; but
they lack many of the essential elements of time-Series trends. making
the handling of such trends by "removing" them from the data, or by
considering them as variables, as with time—series trends, a difficult
and often dubious procedure.

dv" naL.

as

Multiple The methods of multiple correlation an-
Correlation alysis are not generally set forth in much deb
Mechanics* tail in the textbooks, and hence need to be

discusses in this handbook. The working out

of multiple correlation problems looks and
sounds hard and puzzling. In reality it is only long and tedious. Or-
dinarily one can gain a fair comprehension of statistical technique
merely by studying statistical methods. Because of its considerable
detail, however, multiple correlation can be learned only by working
through all the details of it and discovering the meaning of the re-
sults. Many workers using such analysis soon find themselves immersed
in these details and abandon their attempt with the firm conviction
that multiple correlation is too "complicated" or ”technical" for
their purposes. Perhaps it is; but the principles involved in it are
few and simple. Once a knowledge of these principles is ingrained,
the details of the mechanics fall into their prOper categories and
are comprehended without confusion.

The emphasis in this discussion will therefore not be so
much on details as upon a general understanding of procedures and the
reasons for them. Those wanting a complete manual (without reference
to theory) of the arithmetic procedure of carrying through a multiple
correlation solution, with step by step directions and full illustren
tion, including all the shortcuts, will find it in the mimeographed
publication issued by the Bureau of Agricultural Economics, entitled,
"The Use of Punch Card Equipment in Handling Multiple Correlation
Problems." The complete statement of the statistical theory ink
valved with briefer treatment of methods is given in a companion
report, "Correlation Theory and Method Applied to Agricultural Re-
Search".

For our first simple illustrative case, let us take three
series, Ba x and E, the latter being the dependent variable. For
example, u might measure the rainfall, v the sunshine, and x the
yield of crops on a number of different farms. All three series
are first expressed as deviations from their respective averages,
and the deviations set down in three respective columns so that the
three values which go tagether for one farm are on the same line.

The first statistical objective of multiple correlation is to find
two multipliers (more if there er more than two independent variables)--
one for multiplying the values in the u column by, and one for mul~
tiplying the values of v by. These multipliers are the "net regres-
sion coefficients". They may be positive or negative in value, or
one may be positive and the other negative. After the values of u
and v have been multiplied throughout by their respective not re—
rgression coefficients and the products of these multiplications set
down in two new columns, each pair of products, one for each line, is
cross-added and the sums so secured are listed in.a new column headed
x', known as the "regression estimates of x".- The objective is to
find the particular net regression coefficients which will result in

 

* 3y Bradford B. Smith, Cleveland Trust Company

9.,

259

sums (x’ values) most nearly approximating the values of x, or spec-
ifically, h‘ving the highest possible correlation with x. Once these
coefficients have been found, the first_statistica1 objective of mul- “A
tiple correlation-has been obtained,for they define numerically the

not linear relatidnswof x to the other two.

The next statistical objective of multiple correlation is
to determine'actually the Pearsonian coefficient of correlation be-
tween the X values and x':Values, between the "estimates" and the
"actuals". This is about all there is to multiple correlation
processes, per se; the rest is but amplification and development of
ways and means of.atteining these two objectives.

, The net regression coefficients for multiplying the inde—
' pendent variables by in the process of securing the estimates of x and
x' in the preceding, are usually symbblized in statistical literature
by the letter 3, with distinguishing subscripts to designate to which
variable it belongs: gfiu_* b v a x'.

r v

. The coefficient of multiple correlation is designated by
the capital letter R, to distinguish from the simple or gross correl-
ation between two variables, r. ' ' ‘ '

(a) How to find' " ‘ If there were but two forms for which

Values of b. - there were measurements of the values of u,
- v" v, and x, it_would be a simple matter to find
the values of b. Suppose for the first case,
u wasee 4, v was —1 and x was-+2; and for the
second u was —1, v was A 2 and x was +s7. ‘ It would only be necessary

to set up two equations, observation equations, as they are called,
which have two unknowns; and.solve,them.simultaneously:

 

u ‘ _. . V-u .. x

4 b rel b - = ‘ 2
”u “v .+

‘1 b . 2' b7 = 7
‘u ‘v +

Obviously a value for b Vand a value for 2 could be found
u v
which would fit the two equations perfectly. When the left—hand sides
of the equations were then evaluated, which_will be recognized as the
process of determining the val es of x', these results, the regression
estimates, would exactly coincide with the Values of x. There would

be perfect multiple correlation,.v

"‘2 so

 

 

 

ObserVation : u : v z x
number : Terms in b : Terms in b '

1 : 1-4 : ~1 : +-2

2 : —1 : +~2 : + 7

3 : —6 : +3 : +4

4 : 7 : -f1 : -6

5 : i:4 : —5 : —7

Sums - : O : O : 0

Means- : O :‘ O : O

 

In the above little table, the data for three more cases
are shown. Instead of writing out the Eu and theirV for each of

the observation.equations, these symbols are given once at the top.
Here are five observation equations with but two unknows, and we
are confronted with the necessity of finding values for these two
unknowns which will most nearly fit all five equations. 'Obviously
it is not going to be possible to determine the values of Eu and Ev

by the method of solving simultaneously any two of the equations,

for each pair of equations would give different results. Some average
result is what is wanted — and the method for it‘is the Method of
Least Squares.' From these five observation equations, two "normal"

equations are formed, one for each unknown. These normal equations
are then solved simultaneously by any method one cares to use and

the values of 3 determined.

The fonnulation of normal equations follows a definite
sequence, which is well worth memorizing:

(1) Multiply each observation equation by its coefficient of
the first unknown.

(2) Sum the resulting equations, which gives the first normal
equation. _' . ,

(3) _Multiply each observation equation through by its coefficient
of the second unknown and sum the results, which gives the second norm!
equation.

(4) Etc.,if there are other unknowns.

The first unknown in the example is Ru and its coefficient

in the first equation is+~4. Hence the first equation is multiplied
through by f 4. The second equation is similarly multiplied through
by -l, and so on. The sum of the extensions so secured constitutes
the first normal equation.

261

 

+4 (+41 ~11 = +2 ) = (161 -41 = +8“)
u v u V

-1 (“1% +21%, = +7 ) = ( 11% ~2gv ._. —'7 )

-5 (—eg 1, at = 1. 4 ) = (353; ~181)_ = —24 )
u 11 V .

1+7 ( 7911+ 1 _; -6 ) -_- (492ll + 72v : _42 )

-4 (—411 ~5g = :7 ) .= (16}; +209 = +28 )
U. V u

First normal equation......llagl 7”32v e37
The second unknown is 11v and its coefficient in the first

equation is -l. Hence, multiply through by —l, and in analogous fashion
for the other equations.

—1. (+41)u 41g

V

\.

+2) .(-4§u+lgv . 522)

+2 CARI—12% a» +7 ) = (- 21gu +42, = +14 )
+3 (—6pu+3p_v -.- +4 ) ... (—1813L1 fegv = +12 )
+1 (+72u+1p_v = ~6 ) = (+ 71111 +112V = — 6 )
—5 (~43u ~5gv , :7 ) .. (+203u +25%; 3 +35 )

 

+53

Second normal equation. . . . . .. st. 3]; +40h
u v

When these two equations are solved simultaneously, Eu is
found to equal -.3478, and Ev to equal 1.3509, which are the two

net regression coefficients which were to be found.

A moment‘s consideration of these normal equations shows
that the coefficient of b in the first one, 118, is nothing but the

sum of the squares of the u values. The coefficient of};If 'in this

first equation. 4-3, is the sum of the products of corresponding
values of u and v, which is identical with the coefficient of bu in the second

ehation. Some duplicate work_ has therfone been done -- to .‘be avoided
in the future. The coefficient of bv .in the second normal is the

sum of the squares of the v values. vThe terms on the right of the
equality signs in the normal equations are in one ease the sum of the
products of corresponding values of u and x, and in the other of v

262

and x. The normal equations might therefore be expressed algebraic—
ally:

; .Epu 8112) i— €31". gm) z. 8.1.)
. uv / 2 ‘ v
H 20 Q 7 T t z“) 6 x)

The arithmetic values sought for use in the normal equa-

tions are thus gin“), éfiux) , éévz) , éflvx). A systematic scheme
for securing these is illustrate in the following table:

 

Observation.

 

 

 

: Data : Extensions :

numbers : u v x : U ux ux r v vx : x2
1 :T4 -1 'T2’ :+16 —4 8 :1—1 7 2:#’4

2 : —1 +2 +7 :+ l - 2 —7 :fi-4 f14:+49

3 : —6 -+3 'T4 :-+35 ~18 -24 : fi—Q -fl2:-f16

4 :1)? 1-1 -6 :+49 1- 7 ~42 : +1 - 6:146

5 : ~4 -5 ,7 :i—lG 1&20 2 : —f55:«r49
Sums : O O O :illB 1'3 —37 :~+4O -f53: 154

If there were more independent variables than u and v,
there would need to be more columns, one for the square of each
of these variables, and others for all combinations of it with the
other Variables. A column is also given for the squares of x.
When it comes to correlating x with x‘ to determine the coefficient
of multiple correlation, it will be necessary to compute the stan-
dard deviation of x and the necessary squaring of the deviations
might as well be done at this time.

(b) Finding the Coefficient To determine the x' values,
of Multiple Correlation the regression estimates of x, it
(Preliminary) . - will be remembered_that u and v

are to be multiplied by their re—

spective regression coefficients
and associated products added together. This is illustrated in
first part of the following table:

 

 

 

Numbers : —.34'78u £14.35on E Sum Q x :Resid— E .23 Q xx' :"x'2
z : : g x' : :uel. : z :
: _ r : : :x—x ~Z : z :

1 E ~1.39 S «1.35 $-2.74 9+2 i 4.74 5 22.47 E — 5.485 7.51

2 z .35 : 2.70 ”3.05 g+7 : 3.95 : 15.60 : 21.35; 9.30

3 : 2.08 : 4.05 :+6.13 : +4 : ~2.13 : 4.54 : 24.52:37.58

,4 : —2.43 : +1.35 :—1.08 : -6': -4.92 : 24.21 2 + 6.48: 1.17

5 : -+1.39 : —6.75 :—5.36 : -7 : -1.64 : 2.69 : +»37.52:28.29

Sums E o E 0 Q o E o E o 5 69.42 E +84.39E84.29

 

263 '_

The table/ghggs the differences between,theiactual and es-
timated values of x, x - x', designated by the letter Z, and called
"residuals". The standard deviation of Z is the standard error of
estimating x from u and v by the,method of least squares. The cri-
terion of least squares is that this standard deviation of Z shall be
a minimum. The sum of the residuals is naturally zero, since the
residuals are differences between two series which sum to zero: 1 by
definition, and x‘ since it is obtained through multiplying by sons
stants two other series summing to zero by definition, u and v. Since
the sum of the residuals (and consequently the mean) is zero, the ac-
tual items are deviations from mean and the standard deviation may be
quickly determined by taking the mean square of them and then finding
the root. The residuals are here mentioned and their computation
shown because they become highly important in curvilinear multiple
correlation processes, and because the standard error of estimate is
the best single measure of how good a forecasting formula, or regres-
sion equation, actually is.

The coefficient of multiple correlation, it is remembered, '
is but the ordinary Correlation coefficient of x and x'. To determine
it, three other values are needed, the standard deviations of x and x',
and the product moment of x with x'. The products of x and x' and the
squares of x' are shown in the two last columns of the preceding table.
Note that their sums are practically identical. They would be exactly
identical if the arithmetic throughout had been carried far enough.

The value of 20:2) taken from an earlier table is 154. r , is ac-
xx
cordingiy .

84/n
\3? 84/ n) (1154/11)

Since n will cancel out (n is the number of items, 5 in this case)

r a 84 .___.,__,__. - $.- = '74
xx' {Elf—(154 154

The coefficient of multiple correlation in this case .74,
usually symbelized by the letter, R, is commonly written with sub.
scripts to specify the variables involved, the dependent variable'
being given first, and separated from the others by a dot: thus —
Rx.uv I 0.74. This coefficient is always positive in sin. and

ranges from O to 1.00 in value.

Another interesting thing to note at this time is that the
sum of the squared residuals, 69.51 in this case, rounded-to 70, plus
the sum of the squared estimates, equals the sum of the squares of x;
i. e., 70 #34 =- 154, "

‘In the preceding pages has been given all that is actually
necessary to>the working out of a problem in multiple correlation.
The dexcription of method has followed the meaning of correlation as
clasely as possible. In actual practise, short—cuts are used which

264

reduce the labor to a fraction. It is advisable, however, for those}
who have never worked out a multiple'correlation problem before to go
through the steps Just outlined in order to understand the fundamental
nature of the Operations. It will also give them a more than healthy
respect for the shortcuts described in the remainder of this chapter,-
and stimulate their desire to master them.

(c) Labor Saving There are six important way of reducing
Methods. . the labor of multiple correlation computations,

as follows: (1) we have assumed hitherto that

the original data were first expressed as de—

viations from their respective means. The
process of securing these deviations - often a lengthy one - can be
eliminated entirely. ‘

(2) The difficulty of multiplying in securing the extensions
of the original data.may be greatly reduced by simplifying the original
data. This process is called "coding".

(3) In combination with coded data, punch card tabulating
equipment may be employed to reduce the number of multiplications to
secure any product sum to approximately 20 or 30. This is perhaps
the greatest labor saver of all, especially when the number of obser-
vations is over a hundred.

(4) The late discovery of early errors and the labor econ-
sequently entailed in repeating all subsequent computations can be
prevented by introducing at the beginning of the process a check sum
which will check the arithmetic of all steps through the solution of_
the normal equations. ' 1‘

, (5) Owing to the symmetrical nature of the normal equations,
a specific type of solution for them will result in a minimum amount of
labor. ' This is the Doolittle method.

(6) It has been assumed heretofore that it was necessary
to compute the values of x' in order to correlate them with x to
secure the multiple correlation coefficient. To determine R, this
whole process of computing the estimates may be bridged over and
eliminated entirely.

EliminatingiDevigtions from Averggg.

A consideration of the coefficients of the unknowns_1n the
normal equations shows that these are all of the form.€L(u2) orijuv)
are deviations from their respective means. If the original-series.
are denoted by capital letters, U and V, the sum of the squared de—
viations, (u3), may be obtained by squaring and obtaining the sum
of the original values, (U3), and subtracting therefrom.the product
of the mean and the sum of the original series. ,'

phat is, {(113) =_ gm?) .. Mu in

L .

265
If a form were set up it would be something like this:

 

 

U U2
6 36
4 1e
8 64
2 4
Sums O 120
Mean 5
Subtract Mean times Sum (5 x20)..,1oo
(uz) 20

This gives the same result as actually finding the deviations and
summing their squares as below, but with less labor.

 

 

U - M U~M 113
. = u .

e. 5 , +1 1

4, 5 _1 1

8’ 5 +3 9

2' 5 -3 9
Sums 20 .. o 20
Mean 5

A corresponding plan secures product sums like éibrv) withp
out first finding the deviations, by u.eing the formula:

fin”) = .EXUV) — Mu_gy

The product sum of deviations from average may be determined
by multiplying the original two series together, summing, and subtract-
ing the product of the mean of either one times the sum of the other

original series. A form of computation and its comparison might look
like this: . . \

 

 

U V ‘UV ) (U - 5 V — 6 uv
) U“ u =. V

6 8 , 48 ) ( +1 +2 +2

4 4 - 16 ) of_ ( -1 -2 +2

8 3 24 ) with ( +3 —3 -9

2 9 18 ) , ( -3 +3 —9
Sums 20 24 106 . O ' O —14
Means 5 6

-(5'24) or (6.30) Coco-0.120
{(uv)oclccauop-l4
5" 1’

866

Incidentally it should be pointed out that this is ap-
propriate method for the computation of standard deviations and pro-'
duct moments for simple correlation coefficients.

Following is a form set up for recording the computations
of all the sums of squared deviations and product sums needed for a
multiple correlation problem with independent variables A,B,.....N.
and dependent variable X.

Form for preparation of normal equationsr

 

 

 

 

 

 

 

 

 

 

 

‘Explanation :Line ' A g B : : N : x d Check sum
3 3' i f i i ‘ S
Sums : ‘givA ; é;B : nEN : EEJC 1 2i 5
. , . . . ‘ . I
Me ans . i la . Mb : . Mn . MI 3 MS
Extension of A :apl at?) his) ‘ lat-m) :' aux) ‘g 205°)
n Ma times sums :a—2 gMaEA .M‘fiB : main : Mafix “ MpiN
, Line (a~1)-(a-2):a—3 3235) 3-1372) : Him): item) alas}
Extension of B :o_1 I! z (33‘) : : {ll-I): (BX) (BS)
"Mb times sums :b~2 1! as B : :1: N : M x I M S
Line (b—l) - (b—z):b..s : : (b2) : : 13m): 18m) . has)
' : i : ‘ : : :
; ! ; ; ; .
Extension of N :n—l i : : F‘(Nz :Ei05{) SL(NS)
V"Mn times sums :n—2 ' : : :MfiZQ : Ngix Mn£§
Line (n—l)~(n—2) :n-S : L7 :fijn‘): €L(§x) attic)
Extension of X :x—l : : : : 2;{A~) éJXS
" Mx times sums :x-Z l‘ : : : 3 Mfiix Msifi
Line (xel)-(x-2) :x—3 { : : : : 21X?) ELKXS)

 

 

 

This form is explained in detail in the previously mentioned
publications, but is really fairly self-explanatory in the light of what
has just been discussed. When each space has been filled in with the
data designated by the symbols, the various values for the normal equa-
tions have been determined without recourse to the actual computations
of deviations from average! -

The meaning of the check sum column will be discuSSed later.

Coding.

Even when product sums and summed squares are computed by
the method outlined in the preceding section, there is ordinarily a
great deal of tedious multiplication work to be accomplished. To
obtain -4Xuv), for example, it is necessary to multiply each value of
U by the corresponding value of V. This multiplication labor can
often be reduced by "coding" the original values. ‘ Thus, if the
values of U range from 1024 to 1088, from each figure may be sub—
tracted 1020; the series would then range from 4 to 68. The results
as far as the correlation is concerned are identical, for correlation
is concerned with variation from average value. Adding or subtract—
ing a constant from a series in no wise affects its correlation with
or regression on another series. The correlation and regression co-
efficients are unchanged by the process.

Following the same principle, the series can be still fur-
ther reduced by dividing by 3 and dropping the fractions. The series
will then range from 1 to 22. This will not affect the correlation
coefficient materially, but the regression coefficient will be raised
three times, which must be remembered when the final results are "de-
coded" and prepared for presentation. Dropping the fractions in-
troduces an error which. when the divisor is 3, may be as much as l
_and l/2 units of the original series, or .5 of a unit of the coded
series; but ordinarily if the coding is such as to include nine-tenths
of the original items in a range of 20 to 30 coded units, no signifi-
cant differences Will result.

Hence the simplification of series by coding prior to cor-
relating will introduce significant errors only when there is either
high intercorrelation between the independents or an exceedingly high
multiple correlation coefficient. Coding is therefore recommended as
one of the greatest labor savers in multiple correlation work. It
speeds up the work tremendously when punch card equipment is employed
as described in the next section; but it also helps greatly where
computing machines alone are employed. It is indispensable in those
laboratories where no multiplying machines are used, the clerks quickly
learning their multiplication tables up to 20 or 30.

When the series have all been coded, correlation processes
proceed precisely as they-would were the original series employed. The
fact that they hays been coded may be forgotten entirely until the very
last thing when the results are "do-coded" for presentation purposes.

Punch-Card Eguippent.

Punch-card equipment may be used to expedite the computation
of the values to put in the normal equations; specifically, to secure
“the.figpres that go on the lines of designation ending in "1" in the

368 ‘ 3

fonn given earlier. Such eouipment begins to says time as soon as
there are a hundred observations; and sayes it at an increasing rate
as the number of observations increase. One card for each observa-
tion is punched with the coded data for the several variables in their
designated locations on the card. The principle involved in this re-
duction of labor is quite simple: Suppose that we are attempting to
secure the sum of_the squares of the A items. Undoubtedly there is

a large number of items in the A series with like values. It is
easier to square a given value once and then multiply it by the num-
ber of times it appears, than to square it over and over again and
add the results. This is what the punch card enables us to do, The
cards are sorted in the sorting machine into as many packs as there
are values of A. There will be, Say 30 packs, if the data have been
coded. In the first pack, the value of'A on each card will be 1, in
the second 2, etc. we have but to square the value of A and multiply
it by the number of cards in the pack, for each pack, and then add the
products so secured,to obtain the Sum of the squares of the A values.
Or more simply yet, the cards in each pack may be taken over to the
tabulating machine and the A values summed, and the sum so secured
multiplied by the value of A for the given pack to get the same set
of products to be added together.

While the cards are in packs of constant A value, we may
also add the B values for each pack. These sums are then multiplied
by the associated A value and the products added together to give (A3).
This process is but the grouping together of all the E values which
have a constant value of A associated with them; and then instead of
multiplying each 3 value separately by the A value and adding, the B
' values are added together first and then multiplied by the given A
value-~a perfectly legitimate process as a moment's consideration will
show.

By this same process, all the extensions can be quickly ac-
complished. The labor of weeks can be reduced to a matter of hours.
Anyone who undertakes this type of task will have little difficulty in
working out convenient forms and systematic procedures; or one may refer
to the publication mentioned earlier in this chapter and there find the
forms already prepared. .

The Gheck SumI

_After the data have been coded and listed down in columns,
side by side, a final column should be added, headed "Check Sum", or
"S". 'The value listed in this column on any particular line is the
sum of the independent values and the dependent which are listed on

239

_ the same line. For example,_ in the illustration below. 21 3+8 + 8 + 7.

v U v x Cheek Sum
S

21

14

19

13

67
16.8

‘ .

v

Sum

Means

(”BNCDIhOI
mpomusoo
.cnozmoo'mxz
G)

. From the time of its first listing onward, the check sum is treated as
though it,were one of the Variables in the problem except that its stan-
dard deviation need not be computed. The first function of the check
sum is tooheck the addition of the series. for the sum of the sums of
the series should equal the sum of the check sums: 30 + 24 + 23 8 6'7.

The next function is to check the computation of the means:
5 + 5 + 5.8 '5 16. 8. If the sums. howeverJ ar divided he wron
number of items. the check sum will not show it. Care should therefore
be exercised at this point. ' 1

The next function of the check sum is to check all the ex-
tensions. This is done by mltiplying U. for‘example, by the cheek. sum
awhen it is multiplied by itself and the other variables. as in the'
following:- . '

U2 uv UX us

 

 

U v x s z : v2 yr vs :etc.

5 e 7 21 : 35 49 42 .125 : e4 . .56 '168 :

4 4 e 14 : 15 1s 24 56 1 16 - 24 55 :

e 3” 8'., 19 4 54 24 54 .152 : 9 24 57 :

2 9 2 13 z 4 ‘ 18 ‘4 26 : 91 18 117 :
Sums.......... : :170 . 122. 399 x

120 106 134 360 '

Then the sum of the products of any U value times itself and the cor-
responding values of the other variables” is equa‘ to..U timesthe choc}?
sum. for this sum is only the sum of these veryivariables. Thus,

36 + 48 + 42 = 126.- When the extensions are added, the totals
should check, therefore, and the extensions are thus proved:

120 «r 106 +_ 134 a 360. ’

‘When V is extended it is necessary to go back and pick up
a previous extension to check to the VS. total. namely. £(Uv). Thus

me+4mo+1% - we
_ (um +fiv4)+ggvx)= {<VS)

,; -- ,.- ‘ a ,1

as ‘ "f; -

v. . ; " — g l‘. V
The rule is. simple: ..The..prod:uc.t.va-fi-m~<veri%le-with thepheck~~ -.......--..~
sum should equal the sum of the totals of the extensions of that vari-
ables (and itself).

The recording of these check sums can be systematized by
carrying them in a column as shown in the form given above. Studying
this form a minute will show that according to the rule just stated
the items which add to the check sum are on a line (the "line of
check") which starts at the head of a column. runs down to the diag-
onal term and then across to the right.

The check sum may also be used to check the items on lines
of designation ending in "2" in the form. Just referred to. by comput-
ing the values indicated on these lines in the S column. The same
"line of check” should be followed. And. the same is true of the "3"
lines. which five the data to go'intogthe normal equations.

Solution of Normal Eouations.

Once the normal equations have been fbund any method of
solving them sinultaneously to determine the value of the b's may be
employed but the Doolittle method is the "most expeditious. It is
worth learning if much work of this type is .to be done, . The check
sum figures described in the preceding section carry right on through
and check the solution. There is a minimum amount of labor required.
Since any method of solution is valid, however, and since the. Doolittle
method has been described in several publications, space will not be
given to its description here. The reader“i_s~_3e~ferxed._to.mio£‘_thé,,__mu .-
following: '

3" : _' . _: c;

..,‘---'

Adams, 0. S. Geodesy-~Application of the Theory of Least
Squares to the Adjustment of Triangulation. 1915. U. S. Coast
and Geodetic Survey. Spec. Pub. 28. .

Wright, T. S. and Hayford, J. F. Adjustment of Cbservations
by Method of Least Scluares with Applications to Geodetic Work.
3d. ed. N.Y.. 1906.

F. C. Mills.--Shtistical Methods, pp. 578-581. Jour. Am.
Statistical Assoc” Dec..1927, pp. 497-500. ~ .

Computing Multiple Correlation Coefficient.

Attention was called to the fact that 83:12) =8::::!) and
hence that RB = €(xx‘)/ fire)..- Both £(x5c') and 2(x'3) are equal
to the product sums of the independent variables with the dependent
times their respective regression coefficients added together. In
the same example referred t',o £(xx') a 313:2) e b 118m!) 2b (vx).

271

.\H\‘. , . ~

Because of this characteristic, the<correlation coefftcienb

-may be computed quickly without first determining the values of x'.
Thus, for this example: -

3211 ((ux) - ‘(;‘;"3478)X(_37) 7 flags
bw {(vX) z: (+1.3509)(+53) -—-- a 3:31.60

84.47.

J!

{(chl) = a x12)

Hence, by this method, Rm“. =' 484/154 . .74

The generalized formula for the computation of R2 may be

/

t given:

32 be .gfiax) + pigbx) +2c {(cx)+++ influx,
1.8 n11 Z112)

 

Once the normal equations hays been solved, all the data
necessary to evaluate this formula have been determined, and it is but,
a moment's work to compute R, a considerably lees laborious plan than
to work out the x' values and then correlate with x in the ordinary
way.

Those who wish to study the algebraic demonstration of the
validity of computing R in this fashion are referred to pages 51-55
of "Correlation Theory and Method Applied to Agricultural Research."

h. Inference from Results of Statistical Analysis, . ’

The term inference is applied in two different ways in sta-
tistical analysis: one, to the statements that can be made as to the
universe from which the sample is taken — as to the whole area or period
or group of persons or objects of which the sample is a part and which
it is intended to represent; the other, to statements that can be made
about some other more or less similar universe - area, period or group -
not included in the sample. To the first of these, the term "statistical
inference" is commonly employed. Suppose the population of a county is
sampled - how good a description of the 1790 families in it are the av-
erages and coefficients for the 325 families studied? This is sta—
tistical inference. How true are they for any other county? - This is

"“theiotherlkind of inference - sometimes called "logical" or "inductive".

The'distinction, however is by no means as clearucut as the above would

suggest. Suppose the county has been taken as "representative" QiJNflLH w M,,

part of a state? or is one of several counties taken as representative

of a region? "Statistical inference" will now include many statements
belonging in the second category under the first assumption. Obviously
the kind of inference depends technically upon the definition and scdpeh

 

 

1‘ ~--‘

,1:
I

‘—~

27?

of the universe and the sampling procedure. In logic and in fact, “ -
the inference is the same regardless of how named. The truth of the
- foregoing is even more apparent in timenseries analysis. Often all
the available data are included in the analysis. One may say if
one wishes that no sampling is done in such a case, and hence the
whole problem is one of inductive inference as to some future, or
prior, period. Or instead, one may say that the period taken is
a sample out of the recent past, and use the term statistical in—
“ ference for the process of making statements that would describe
'“t 9 recent past. In any case, the forecasting statements represent
in vs or logical inference.

The problem of inference arises with respect to averages
“nd other summary expressions for single attributes, and also with
respect to coefficients of correlation and regression as measures .
of relation between attributes or variables. Most of the discussion\ ,.¢5~~
in this section will relate to the latter. But some further dis- -
-cusaion of the former is needed as an introduction to the latter.
Pure statistical inferences as to a universe derived from a sample
of it have varying validity, it has already been pointed out, ac-
cording to the relative size of the probable error of the mean, and.
according to how clearly the data themselves are distributed accord-
ing to pure chance. Given disturbing influences in the data, such
as geographical factors, which produce departures from constant prob-
ability, and the basis for statistical inference is seriously impaired.
The mathematicians have devised various methods of comparing the dis-
persion of statistical ratios with certain theoretical norms. The
Lexis ratio uses as norm the conventional standard deviation computed ~~i~r -
on the assumption of binomial (Bernouilli) distribution. The pro-
cedure for determining a Lexis ratio is to break the data up into sub—
series on some rational basis - smaller geographic units, for example,
if geographic data are being used ~ and compute the standard deviation
of the means of the sub-series. This divided by tho Bernouilli stan—
dard deviation gives the Lexis ratio. If it is greater than 1, dis-
turbing factors are present producing departures from constant prob-
ability. Often indeed is it greater than 1. The most available
discussion of this subject is in Rietz's Handbook of Mathematicol
'Statistics, Chapter VI. Whatever inferences are drawn should take
account of this or some similar ratio. See the discussion of this by
Professor E. B. Wilson in the Journal of the American Statistical Asso-
ciation, June 1927, pp. 209~12. Earlier analysis may have taught us
that a Lexis ratio of 5 is to be expected with etch data. In other
cases, we may be in the position of having Ito gambTe on the typical-
ness of our realized experience" (E. B. Wilson, p. 221).

Inference £33m Correlaticn83*

(1) Statistical It should be clear from the discussion
Inference which has gone before that the regression
equation and the correlation coefficient are
_ ‘. merely statements of numerical relationships
between certain numbers which have been used. Of themselves, the
correlation coefficient and the regression equation mean nothing what-
ever except with regard to the particular set of numbers which haven

*.Elmer 3. Working, University of Minnesota.

,273

been used to obtain them. For any further meaning, we must~look'to’x
the original date end sec Whether the relationships they exhibit have"
any significance with regard to a larger body of the date.

Suppose we have a universe of one thousand objects or in-
dividuals. This universe exhibits certain characteristics or veriates.
Now suppose further that we want to find how two of these varietes are
related in this universe. Do we need to correlate these for every one
of the thousand objects? Hot as a rule. We can correlate a sample,
say one hundred, of the observations and if that snxple is representative,
the results will be virtually identical with the results which would have
been obtained by using the entire one thousand. In order that the sample
be representative, every sort of relationship existing in the given un-
iverse must he found in the sample, and the occurences of these relation-
ships must be in the same proportion in the sample as in the universe.

In case the universe is homogeneous, e,reprcsentotive sample will be

get if the observations are selected at random. In case the universe

is not homogeneous, random or ”simple" sampling cannot be depended upon
to yield a representative sample and other means must be employed if
possible. It is to be noted, however, that these other means of repre-
sentative sampling me' be employed only on the basis of some knowledge
of the entire universe. Furthermore, they do not obviate the necessity
of observations from all parts of the universe being included if the
sample is to be representative.

Suppose for example we are interested in the relationship be-
tween the density ef population and the amount of precipitation, both
of which vary geographically. If we are concerned only with conditions
in the United States, we may take a random sample of 100 counties over
the whole country. Subject to certain limitations, the relationships
found to exist in this sample may be expected to be the some as those of
the entire United States. This extension of the results of the correla—
tion is in no manner dependent upon the methodology of correlation, but
entirely upon the nature of the data and the method of toking the sample.
If we had taken the 104 counties in the 4 states Arizona, Idaho, Nevada
and Utah, we could not expect the relationships found to apply to the
entire United States. Similarly we could not expect the relationships
existing in the 99 counties of Iowa to hold true for the entire Uhited
States. The results of a correlation which includes a contiguous area
must never be supposed to hold true for any larger area unless there is
reason for believing that the conditions of the smaller area are repre~
sentative of the larger.

The effect of selectivity in the smnpling must always be cen~
sidered carefully in its relation to correlation results. The question-
naire and the survey method, as already pointed out, are nearly always
selective. Suppose the problem is one of the relation between the
quantity of fertilizer applied to a crop and the yield. This relation—
ship is usually ourvilinesr, the curve flattening out with the heayier
applications. It is apparent that if correlation analysis is to re-
veal this relationship, the data analyzed must cover the whole range
from complete lack of fertilizer to the point where additional applica~
tions do not result in any further increase in yield. If the sample

274

included only low applications of fertilizer, the results of analysis
would show marked increases in yield accompanying each increase in
quantity of fertilizer; 1.9., the regression line or curve would be ‘
comparatively steep, and the conclusion might be that there is a close
relationship between quantity of fertilizer and yield. On the other
hand, of the data all showed Large applications of fertilizer, the

r sults would show much less increase in yield for each increase in
quantity of fertilizer, the regression curve would be comparatively
flat, and the conclusion might be that there is little relation be-
tween quantity of fertilizer and yield.*

The converse of the foregoing is that sometimes a selected
sample is what is wanted. For example, availobbo data from earlier
studies may not have provided enough.detail to enable one to locate
the point of diminishing average and total returns; hence the fields
taken in the present survey may need to be mostly those receiving
heavy applications of fertilizer. If the problem is one involving
the use of increasing mnmints of hired labor, only the larger forms
will be taken. It is entirely proper to define the universe of any
study in such a way as to isolate the cases in which the desired
relationship is to be observed.

A sample which includes all the conditions in exactly the
same proportions in which they occur in the universe may show so
few cases at the extremes that only a very uncertain picture is re-
vealed ef the relationships at the extremes. The five farms using
light applications of fertilizer may be abnormal in respects~notw
covered in any of the measured variables; hence the shape of the
acmial curve obtained at this point'may be highly misleading. The
conclusion to be drawn from this is that if the rule of identical
proportions in the sample and the universe above stated is to be
adhered to, then a much larger sample is needed then any of the
tests of size of sample would suggest, especially if the relation—
ship is in the least_eurvilinear at the extremes. 'Even with linear
relationships, the Pearsonian analysis so exaggerates extremes that
a few abnormal cases at the ends may shift the regression lines out
of all proportion to reason. The criterion in such cases would
rather seem to be enough cases for a sample at each of the important
sections of the regression line or curve. Another possible solution,
but fraught with considerable hazards, is loading the sample with an
abnormal proportion of cases toward the two extremes. With relation~
ships stated in linear form, this may have the effect of producing
a regression line which reflects unduly the conditions at the extremes.
This may be avoided by using the excess of extreme cases only in a
supplementary mibsequent analysis, which may reveal evidence of curv-

ilinearity.

In studying time series the situation is similar save that
there is seldom occasion to select a sample from a given universe in
order to reduce the number of observations correlated and hence the
laboriousness of the calculations. If we are concerned only with a

 

*This and the two following paragraphs inserted by the Committee.

2%

period of time for which data are available, the number of observa~
tions for the entire universe is likely to be small enough so that
the entire period may be included in the study. A case to the con-
trary would be the following: Suppose, it were desired to study the
co-variation of the volume of trading in wheat futures and the range
in price of wheat futures. If daily data were available over a
period of 7 years and if we wished to study the data by days rather
than by months, we would find that to correlate about 2100 observa-
tions was too laborious. we would then select at random a sample

of these data. If this was properly done we could feel assured

that the results would approximate what we would have obtained had
we correlated the entire 2100 obscrvations. Other objectives would
be better served by forming the individual observations into monthly
or yearly averages or aggregates. '

Ordinarily in the study of time series something more is
wanted than a description of the co-variation of available data. It
is desired to $§dge of the future from the relationships of the past.
Thus we may study the relationship between stock prices and interest
rates in the period 1900-1928. hoping that we may learn what will be
the relationship between them in 1939, 1930 and other years of the
future. With regard to this problem, the "statistical meaning" of
correlation can tell us nothing definite. We have no assurance
from the theory of sampling that the relationships of a sample will
hold true beyond the bounds of the group of data from which the
sample was selected. A continuous time series may be looked upon
in either of two ways——as a universe for which we have complete data,
or as a sample which is not in the least random and hence may not
be representative of a longer period of time. In neither case does
the theory of sampling as such warrant us in making any statement to
the effect that the conditions of co—variatien obtaining in the sample
will hold true at any time in the future. Thus if a study were made
of the relation between the price of wheat in Minneapolis and its pro-
tein content, and the data usad constituted a random sample of all the
cars sold on the market during the year 1937—28, one would be warranted
in Saying that the results applied to all wheat sold in Minneapolis in
1927-28, but not in applying theSe results to any other year. For the
extension of the meaning beyond the bounds of the particular universe
.from which the representative sample is taken, we must look to other
reasons.

‘Inductive - I Although the theory of sampling does not
Inference warrant the extension of the results of cor—

relation to a period of time in the future or
_ to a region beyond the bounds of that which we
have studies, we are not to assume from this
that our experience of the past can tell us nothing of the future, or
that experience gained concerning conditions in one country can tell
us nothing of what to expect in another. But we must be eXceedingly
careful how we try to judge the unknown from the known. We must examine
very carefully just how the past may tell us of the future, or condi—
tions in one region of conditions in another region. Since the problem

2fi6

is virtually the Same for both time and geographic series, and since
the problem is more frequently met in time series, we may well con—
centrate our attention upon it. Let us examine, then, the grounds
upon which people have forecast the future from the past. The followa
ing expresses the kind of hypothesis upon which much of the recent
price forecasting has been based.

"If in past years we find that certain factors as measured
in March have borne systematic relationships to prices in the next
December, with great regularity, is it not reasonable to suppose that
by applying those relationships to the sam factors in a current year,
it will be possible to forecast the price Often such an hypothesis
is implied but not explicitly stated, it apparently being assumed that
the truth of the hypothesis cannot reasonably be questioned.

But why, we may ask, should relationships which have existed
in the past be expected to remain the same in the-future? To such a
question it may be answered that the expectation is based upon the uni— “‘
formity of nature-~that it is no different from the expectation that
an increase in the temperature of a certain amount of oxygen from 20
degrees to 30 degrees will always_be accompaniedfifihe same increase in
pressure if the volume remains the same. Without such uniformity of
nature, it may be said, everything would be chaotic: there could be
no knowledge.

Here we must be on our guard. The uniformity of nature in
which we believe is a uniformity of cause and effectu But what is
measured by correlation is co-variation, not cause and effect. Cor—
relation analyzes and describes co—variation, not the nature of cause
and effect between the variables. If we are to-be assured that the
relationships of co—variation existing in the past will continue in
the future, we must show that the relationships are those of cause
and effect as well as of co-variation. We must examine, then, how
causation may be related to co-variation.**

Let us use as an example a case where the variables are few
and the cause-and—effect relationships both simple and known. Such a
case is that of the pressure, volume and temperature of any given quan-
tity of a gas. The relationship between these is such that PV = RT.***

 

* B.B. Smith, "Forecasting the Vblume and Value of the Cotton Crop".
Journal of the American Statistical Association, V01. XXII, p.443
(December 1927) ‘ '

** These statements are to be interpreted with all the qualifications
named in Volume I in the section "What is Science?" Cause as here
used is to be understood merely as a relativel high degree of rob-
ability, not as absolute necessity. The searc for causes is t e
search for antecedent probabilities; .

*** Where P represents the pressure, V the volume, T the "absolute"
temperature and R a constant which depends upon the amount of the
gas and upon the units in which the variables are measured. While
this exact relationship breaks down under extremes of pressure and
temperature, it may for all ordinary ranges be considered to be a
universal cause-and—effect relationship.

2%

Suppose, then. that we have the volume and temperature of a given
amount of air under control; varying them and observing the result-
ant pressure. Temperature in this case unequivocally a cause of
pressure; and so is volume. Not only that, but the relationships
proved to be curvilinear, and when the logarithms of the two fac-
tors were correlated the coefficient was 1.000. Such a high cor-
relation would seem surely to indicate a virtually complete explana-
tion of the variations in the pressure of this particular amount of
air. But when the relationships found to exist between pressure
and temperature in the first set of obserVations were used to pre-
dict the pressures of the other three sets, instead of an average
error of .00002 in the logarithms, we had errors of .3387, .5965.
and .3228 respectively. The attempt to estimate the pressures

in the other three sets of observations by the use of the regres~
sion of pressure on temperature as found for any 9ne of the four
sets gave similarly large errors. This is evident from a compar-
ison of the regressions obtained‘from each of the four sets.; They
were: “

(1) log I“ ._ ~7.3214 {—3.4485 log 'I! V. .
(11) log P‘ . 4.224 +- .9799 log T
(111) log P' a f .3177 .4188 log T

(IV) log P' 314.4554 — .1056 log T

The coefficients of correlation between the logarithms of
pressure and temperature were as.fo110ws:

1. 1.0000
II. .0126
III. . 7509
Iv. . 5643

These varying results were obtained despite the fact that .
the causal relationship of pressure and temperature of this quantity '
of air was the same throughout all four sets of observations.

The difficulty is of course due to the failure to include
a third variable - volume. The third variable was needed for a
complete system of cause-and-effect relationships for a given amount
of air.* But just such ommission as we perpetrated here is charac-
teristic of most economic analysis. Most phenomena with which the

 

* By "complete system of cause-and-effect relationships“ it is
not meant that there are no causes which lie back of the vari—
ables, or that there are no effects beyond those upon thelvari» -
ables included. It is meant only that these three have a ' '
definite relationship and that if the values of two of the
variables are determined the remaining variable can have one
and only one value regardless of what may be the magnitude of
any other factors. In the terminology.of mathematics, this is'
a functional relationship such that the dependent variable is'
a function of the several independents.

5'29

economist is concerned have causal connections with a very large
number of other factors, and a complete system of cause-and—effect
relationships might need to include anywhere from a half dozen to
a score or more variables. Multiple correlation technique may
theoretically be extended to include any number of variables. In
practice, seldom more than three or four can be included. The cor-
relation of two variables in the above example includes a larger
part of the "causes" than a correlation of 8 or 4 factors in many
economic problems.

If in our example we correlate all three variables, thus
having a complete system of cause—and—effect relationships, we find
that the regression equations express the causal relationship. The
regression equations are as follows:"I

II log P' = .7050 — 1.0000 10g V 1'1.0000 log T
III logP' _-_- .7084 - .9984 log V +1.0000 log T
IV log P' -_- .7050 - .100000 10g V +1.0000 log ’1‘

These regressions approximate very closely the true cause~and~effect
relationship. Number III is not as close as the other two because

the computations made in obtainingit were not carried out to so ex-
treme a degree of accuracy.

The foregoing illustrates the principal conditions necessary
in order that the regressions may without doubt express the true cause-
and-effect relationships. They are: (1) The inclusion of a complete
system and only a complete system of cause-and-effect relationships.
(2) The existence of linear relationships of co-variation between the
variables or of curve—linear relationships which can be precisely de-
termined. (3) Absence of perfect correlation between any two of the
variables.

In addition to these a fourth should be mentioned. It is
that every one of the variables which comprise a complete system of
cause-and~effect relationships must fluctuate in the sample correlated.
If any one does not fluctuate its effect cannot be observed. Stated
in this way, the situation is self-evident and yet it is likely that
one of the principal difficulties encountered in forecasting by re-
greSSion equations is that during the period studied certain important
causal factors are dormant. ‘

 

* It is impossible to obtain a regression equation from the data of
Table I if the logarithms of all three variables are used and the
computations carried out with extreme accuracy. Since the correla»
tion r23 (the correlation between the two independent variables) is
equal to 1.00000, the normal equations cannot be solved. If the work
is not carried out with extreme accuracy and log P +—log V does not
exactly equal log T plus a constant the equations can be solved. The
solutions are without meaning however, because their values depend on
the errors and are very greatly influenced by very slight errors in
the first order 3's. This whole situation is due, of course, to the
perfect correlation which exists between each two of the three varia
ables when air is compressed under adiabatic conditions.

279

The four conditions necessary in order that the mathematical?
process of correlation“may be certain to yield the causal relationships
probably never exist in actual study of economic data. Consequently
there are no grounds for assuming that the correlation analysis will
give us the results we desire unless we have some further evidence.

The problem resolves itself into a matter of Judging, if pOSSible, how
seriously the results.of a correlation are affected by the failure to
meet these four conditions. This is a matter which the merits of each
individual case must decide. A general treatment can only suggest
methods of approach and evaluate the relative importance of the four
conditions.

We need give detailed consideration to the significance of
failure to meet only the first and the fourth of these conditions.
The third condition is a mere technicality for virtually every conceiv-
able problem of actual analysis conforms with its requirements. The
second condition requires proper technique of statistical manipulation
in order that it may be met, and while this may be difficult or even
impossible of attainment, most problems can be dealt with by some one
of the methods in common use. The necessity of constructing a cor-
relation tabls or scatter diagram of the relation between each "inde-
pendent" variable and the "dependent" variable is imperative,.however,
in order that some idea may be gained of the homoscedasticity and hom-
ocliticy of the distributions as well as linearity of regression.

What will be the effect if some of the factors of causal ink
portance are not included? The extent . .g of the effect will de-
pend updn the number of the factors omitted and also upon the relations
of sensations and covariation which those factors have. If none of
the omitted variables is correlated with any one of the independent
variables which are included. the only effect will be to lower the
coefficient of correlation. The regression equation will express the
true causal relation between the dependent variable and those independ-§
ent variables which are included. ‘

- If, on the other hand, there be a correlation between an
omitted factor and one of the independent variables used, the effect,
or a part of the effect, of the omitted variable will be attributed

in the regression equation to that independent variable with which it
is correlated. In case correlation exists between an omitted variable
and more than one of the included independent‘variables, part or the
whole of its causal effect will be attributed in varying degree to

the several independent factors which vary with it. The degree to
which the regression-equation is affected will depend upon the followh
ing conditions: (I) the number of the omitted variables, (2) the im-
portance of their causal effect upon the dependent, and (3) the close-
ness of correlation between the omitted variables and the included
independent variables, '

The way in which an omitted variable has its effect attrib-
'uted to another can be shown most readily by considering a case of
two virilbles where this is true. If for the period 1900 to 1927,
we correlate the deflated yearly average price of cats with a weighted

has

average per-acre yield of potatoes in Iowa and Nebraska, we obtain

a coefficient of correlation of .255. Obviously this correlation
between the yield of potatoes in Iowa and Nebraska and the price of
oats does not mean that the yield of potatoes has an Important effect
on the price of oats of the following season, nor does it mean that
the price of oats for the crop-year has any effect upon the potato
yield of the preceding summer. It is not to be denied, of course,
that there is a causal relationship in this case. weather conditions
affect both the yield of potatoes and the yield of cats. The per
acre yield of cats in turn affects the production and the price of
oats. ‘ .

Indeed it may well be argued that correlation always in—
dicates causal connection. No matter how remote the cause may be
or how devious the path of causation, we may hold that every occur-
rence has a cause and that, in the last analysis, there is no such
thing as pure chance. Though we may say that correlations which
are due to trend have no significance, this does not mean that there
is no causal connection between the variables. Thus, if we were_to
correlate for some period the shipments of pineapples from the Hawiaan
Islands and the production of cantaloupes in the state of California,
we would probably find them positively correlated. Furthermore, such
a correlation would be due to the causal relation between Ehe two -
both depending upon the growth of population, improvement in transporter
tion and refrigerating facilities, changes in habits of consumption, and
other things which change with the passage of time and are causally re-
lated.*

If, however, the causal connection is devious or uncertain,
even a high degree_of correlation is of little significance, particularly
if the relations established are to be used for purposes of prediction.
There are too many chances that the causal connection may be broken or
modified with neither warning nor apparent reason. Thus, given a close
relationship between the yield of potatoes and the price of cats, either
potatoes or oats may become subject to a new disease which will entirely
change conditions. There may be changes in the regions where oats and
potatoes are grown so that their yields are no longer largely dependent
upon the same set of weather conditions. -Methods of culture may change
so as profoundly to affect the amount of oats planted and thereby make
the yield of potatoes an altogether erroneous measure of the cat pro-’
duction. All these and other things would change the relationships
of co-variation even though the ultimate cause-and~effect relations
ships -— that is, the uniformities of nature -~ remain the same.

It is of utmost importance then, if we are to Judge concern,»
ing any universe of data from a sample which is not random, that we
find causal relationships; and in order to do this we must ayoid omitte
ing from the correlation any causally important variable whose variap
tions are correlated with the other "independent" variables -— unless,
indeed there be a direct and certain relationship of causation between
the omitted and the included variables. .

r another view of the reason for such correlation see Tule G.lL

"Why do we Sometimes get Nonsense Correlations Between Time-Seriesfl,
Journal of the Royal Statistical Society, Jan. 1926. ' ' ~ 7

I
v

25

20

I5

 

FIGURE I.

 

 

 

 

 

'" 'L.1mmmmu‘v1m1‘.‘wmnnnmmnmnmmnmmmmmmnmmnnmwwummvm

O '! mmrmmmj

 

 

 

 

WW
‘5 l0

 

I5]

 

 

    

diff/III ’3‘1755' 71117732171 I’WTFILM'ZR” 7117!] \ I

20

 

25

284

1918-21 1923-25
31.28 . .8019 .5965
Ri.23 a .6431 .3558
B12.3 = _.4555 —.5397
313.2 a .6980 .1418
hma- «ma amm
b13.2 a .2574 .1562

Is it reasonable to suppose that these regressions express
causal relationships? let us look first at the net regression of
price on receipts (blz 3).. In both these periods the general nature
of the relationship is consistent with what we should expect to be
the causal connectionr~that is, increases in supply accompany decreases
in price. So far, so good; but it is difficult if not impossible to
judge whether the extent of decrease in prices accompanying a given in-
crease in receipts indicates a reasonable causal relationship.

When we come to the regression of prices on the index of the
general price level, we are in a better position to Judge. The import-
ance of the price level index is that it indicates changes in the pur-
chasing power of money, and insofar as it does this we should expect a
very definite causal relationship. we should expect that if the value
of money doubles, prices will be cut in half, and that (aside from
temporary lags) this would be a virtually universal causal relationship.
In order for the regression to express the causal relationship, then,
the price of butter should exhibit the same per cent increase or de-
crease as the general level of prices. Since in the first period the
average price of butter was 56.2 cents and the average of the index
was 203.2, the average causal relation to be expected is about .27?
as compared with a regression (b13 2) of .257 — a discrepancy which
is not very great. In the later périod with the price of butter emu
eraging 44.4 and the index 154.1 the a priori relation to be expected
is about .288 which compared with a net regression feund to be .1562.
In view of this difference, it would appear that the net regression
of the price of butter on the index does not give the causal relation-
ship.

mhe comparison of the regressions for the two periods shows,
of course, that both regressions cannot express the causal relationship
between the three factors. It indicates that neither of the regressions
'given her. or one obtained from the entire period can be trusted to be
the causal relationship or to give satisfactory results for prediction.*

nthis particular example, of course, the smallness of the coefficient
of corre ati on would also indicate that the results could not be expected
to give accurate estimates but the smallness of the coefficient is due to
keepin down the number of variables for the sake of simplicity and is not
typica, in this respect, of correlations which would be used for prediction,

 

285

In connection with this example it is worth while to note
that the larger coefficient of correlation obtained for the first
period is due primarily to the large variation in the general price
level or value of money. This in turn caused a great variation in
the price of butter—~a variation of a type which we can readily ex-
plain. Hence the high correlation.

The foregoing analysis does not illustrate all the ways
in which it is possible to judge whether or not regression equations
express causal relationships. There are other means and it is well
‘to employ any which may suggest themselves as being adapted to the
particular problem at hand. Whatever the other tests employed, how—
ever, one test ought not to be omitted in any time-series analysis,
namely, the test of the Stability of the data.

It is not to be supposed that the results of correlating
time series and geographic series are altogether useless unless satis-
factory evidence ean be given that the relations of co-variation which
hare been found are also those of cause and effect. Indeed it may be
that the knowledge of certain conditions of co-variation is very useful
even if it is evident that they do not express causal relationships.
Any sort of knowledge of the past may have some value in judging of the
future. Nevertheless it cannot be too strongly insisted that the error
of predictions which are based on observation of past events is ab-
solutely unknown unless the observation of past events has yielded a
knowledge of cause-and-effect relationships.

Though it is necessary to resort to g priori analysis no
matter what the results of the correlation process, this fact by no
means indicates that the correlation analysis is of small importance.
Induction and deduction must be combined in almost every problem, their
proportions depending updn the problem at hand and the nature of the
data available for studying it. 'The correlation analysis can present
evidence of the same general sort as any other inductive reasoning.
The use of multiple correlation methods, it must be remembered, does
not by any means indicate that "all other things have been held con-
stantant". We must have Something more than a description of a set
of numbers -— something more even than the numbers in their entirety
if we are to interpret them. Correlation cannot of itself go back of
the numbers which it describes.

Inference from' If a correlation of certain time series
Correlation in covering a given period yields a reason—
Time Series* ably high correlation coefficient, with

what assurance may a forecast be made?

Attempts to forecast from correlation in time series rest
upon the expectation that the relationships between the independent
variables and the dependent determined for the period of time included
in the correlation will continue to hold for the period of forecast,
and on the further expectation that the difference between the fore-
cast and the actual will approximate the differences between the actual

 

* L. H. Bean, Bureau of Agricultural Economics.

2&6

dependent series and the series computed from the independent Variables
for the period covered by the correlation. These expectations are valid
only in so far as economic phenomena in the immediate future may be re;
lied upon to behave as they have in-the past and in so far as no develOp-
ments arise in the immediate future which-did not exist during the period
covered by the correlation or which were not included as variables in the
correlation. In cases where correlations are based on definitely known
independent causal factors influencing a given dependent variable, the,
problem of inference arises primarily from the fact that these expecta»
tions may at any time be upset by new factors. In cases where correlan
tions are based on questionable causal independent factors which may
merely show co-veriation with the dependent, the expectations are much
less valid if at all reliable. '

The validity of expecting correlation relationships of a
selected period of time to hold also for another period following or
preceding is not enhanced by large numerical values of the Correlation
coefficient and the small values of the errors of estimate. The correl-
ation device as developed to date rests on the theory that coefficients
computed from different samples of observations would give approximately
the same results differing from each other within the narrow limits
governed by the probable error. But segments of time series are not
different samples of a larger body of observations comparable with
samples of data in which time is not an element. This does not mean
that it is not possible to correlate the same set of data for different
periods and obtain coefficients of about the same magnitude for each
period. Such stability when obtained is highly significant, and suggests
causal relationships, but possibly for these periods only. In a constantly
growing and changing economic organization, the results obtained for the
past may or may not be applicable to the future. The future only can
demonstrate to what extent new factors, such as changes in financial or
commercial conditions, habits, production methods, standard of living,
will make correlation results descriptive of a small portion of the
past applicable also to the immediate future. But where the correla—
tion is based on sound economic reasoning, supported by results that have
held good for a reasonably long period, there is a reasonable degree or
assurance that economic relationships even in a dynamic society will not
often abruptly break away from the past. - Where forecasts must be made,
as in the case of laying plans for future business or farming operations,
a description of the past as contained in a sound correlation analysis
is of great service even though it is necessary always to be watchful of
unexpected developments.

With this limitation on forecasts from correlations recognized,
the problem that confronts the forecaster is whether the correlation
study is based on a sound analysis of causal relationships and on a
proper selection and treatment of dates A proper understanding of the
economic csuse-and—effect relationships is a sine Qua pan of correlation.
It is equally important that the data included in the correlation be
treated so as to reveal their prOpcr effect on the independent variable.
For instance, in a correlation between the changes in cotton acreage as
the dependent variable and the price of cotton and the preceding year's
production as the antecedent independent variables high coefficients of
correlation can be obtained for the period 1902-1924 by using price

287

either as an absolute figure or as a percentage change from the pre-

ceding year's price; but in applying the results in a forecast of the .. ,J
1925 acreage, the formula including absolute prices gave good results"
while the other did not. In that year, even though prices had de-

clined, the level of cotton prices was still high enough to induce
producers to expend acreage, a fact which the formula based on prices
changes (first differences) did not adequately represent. (See p. 8,
Technical Bulletin 50, Factors Affecting Cotton Prices, for this
illustration).

In predicting from results of a multiple curvilinear cor—
relation, the process is one of reading values directly from regres-
sion curves. The multiple curvilinear method, it will be recalled,
begins with the computation of the multiple linear regression equa~
tion. The straight line regressions from this equation are then
manipulated graphically, if the residuals, having been plotted about
the linear regression, indicate a curvilinear instead of a linear
relationship; the purpose of this manipulation being to reduce to a
reasonable minimum the differences between the sum of readings from
- the regression curves and the actual dependent variable. The trans—
formation by free—hand methods of a linear into a curvilinear regres-
sion is almost entirely a matter of judgment and facility in graphics,
colored by'a consideration of economic principles. Consequently, the
transformed curvilinear regression curves can not uSually be described
mathematically.

It is from these final curves that forecasts are made by
reading from each separately and summing the several readings. Ex—-
amples of multiple curvilinear regresaion curves may be found on
page 49, Department Bulletin No. 1440. ,

The large element of personal Judgment that enters into the
construction of the multiple curvilinear regression curves raises a
question as to the validity of the high coefficients of correlation
that often result from such studies. Complicated regression curves
so drawn as to fit most of the residuals may take on shapes which
are difficult to explain. For instance, in the first of the examples
referred to above, it is difficult to explain why the price of hogs
should rise eighteen months later following an improvement in the
corn-hog differential from $8.00 to $2.00 and decline after an im-
provement from $2.00 to $1.00 Complicated regression curves might
easily indicate that other factors were missing in the correlation.
Where regression curves are artifically forced into peculiar shapes
in order to secure high coefficients of correlation. the high co-
efficients have little meaning and the readings from these curves
are likely to give very poor forecasts.

The introduction of time as a Variable in multiple correla—
tion permits the fitting of a curve through the residuals plotted
_IeninIt-time, indicating the effect on the dependent variable of fac~
tors which vary uniformly with time. If this regression curve (of
residuals) is a straight line or a simple curve, and actually repre-
sents factord not included in the correlation but which have the
appearance_of a persistent fairly uniform effect on the independent

288

variable, such as might arise from an increasing population. it may be'
projected and included in the forecast.

(4) Predicting from Correlation Pesults.*

Inference from time~series correlation comes to its final
fruition in the task of prediction-- as for example in the task of
predicting price movements of a commodity. Here it comes into comp
petition with the methods of experienced traders - with the technique
of artists who have learned by trial and error and experience. Thus
put to the test, all of the difficulties'of inference pointed out above
have been brought strongly to the fore. The use of the Pearsonian
coefficient as an indicator of the closeness of relationship of the

for various periods of time may be far from sonstant; and that the mag-
nitude of the coefficient may have been largely determined by a few

In one case where war years were used in measuring a price-supply re-
lationship. the effect of those years gave a coefficient of *~.87 for
a thirty-year period, while if the last four years were omitted the
coefficient was -.75. Just because some one has secured a high co—
efficient of correlation does not necessarily prove or disprove that
a high degree of relationship exists.

Nevertheless, a number of statisticians have used the
Pearsonien method of correlation as a basis for prediction with
signal succese. As to whether the measurements obtained by such
analysis are accurate enough, or too sweeping changes in the relation-
ship take place from time to time. or new factors come into the field
to influence prices, only the good Judgment of the investigator can be
relied upon. Just as with many tools, in the hands of the inexperi-
enced, the result may be harmful to its user or others, in the hands
of one skilled in its use it may be a powerful aid.

. The experience of meeting the competition of the traders has
also taught us much that is illuminating about their methods. We have
seen that many of the old-time traders made no-attempts to improve their
Judgments by means of statistical devices other than securing data as
to supply and roughly estimating the probable demand. very often the
question of how much will be taken at a given price is not given any
consideration. Even as late as 1927-28, old established analysts in
the grain trade estimated the amounts of certain grains that Europe
would import on the basis of what had been taken in other years when
prices were 50 per cent lower. The amount actually_taken was about
one-fourth that estimated. But they will go on repeating this error.
as in the past. The more statistically inelined traders resort to

 

3—3; Dr. Hutzel Metzger. U.S. Bureau of Agricultural Economics.

289

graphs to bring before them a picture of past relationships between
prices and the factors which have affected them. On the basis of
these, and a knowledge of the probable actions of others dealing in -
the commodity, they make their sales or-purchases. Since a number
of factors influence prices, the graphic method may become difficult
to follow. The various factors may be interrelated. With each
factor continually changing, the picture easily becomes almost hope-
lessly confused.

. It is therefore not difficult for a careful statistician to
believe that he has considerable to contribute to the task of predict-
ing prices, And yet only a little experience with it also convinces
him that making this contribution is a man—sized undertaking. The
method here chosen to present the nature of the problem and its meni—
fold difficulties is to take a single commodity, corn, and work through
the details of the prediction of its price.*

Some of the pioneer work on price supply relationships in
agricultural commodities was that done by H. L. Moore.** He measured
relationship between the production of corn and the_December price from
”a1866 to 1911, securing a coefficient of .79; He also fitted a third
degree parabola to these data, and had he calculated an "index of cor-
relation" by the formula introduced by Tolley and Ezekiel for the meas-
urement of curvilinear relationships, he would have obtained an index
of about -.95. Yet, if he had attempted to estimate the price in
1927-28 by means of this curve, he would have been far from correct.

If he had divided his data into two or three periods (these might have
overlapped), or into a series of periods by adding one year and dropping
one each time, computing a "moving"correlation coefficient, he would
have found that the closeness of the relationship increased as 1911 was
approached from approximately~70 for the first twenty years to about .90
for the last twenty. A regression equation based on the last twenty
years would have given him mere accurate results for forecasting in

1911 than one based on the entire period.

Corn lends itself admirably to this type of study. The .
price-supply relationships have apparently changed little in the last
thirty years. Such a small preportion of the crop has been exported
up to this time that crops in other countries can be neglected for
correlation purposes. Likewise, very little is imported and existing
tariffs hays hardly been very little effective. Using only United
States production data on a crop that is not so difficult of estima-
tion, we have as accurate a supply series as we can likely find in
any crop. Also during, and for a considerable time previous to the
period studied, corn has been sold on an organized grain exchange
with a futures market where traders have an opportunity to register
their opinions as to what the price should be Definite grades have
been established and quotations are available for the grades in a
given market for a long period.

 

* This analysis was begun in 1925 at the University of Minnesota
under a fellowship grant from the Laura Spelman Memorial Founda-
tion.

** Moore, H. L., Economic Cycles, Their Law and Cause.

Whatever changes are taking p1

ace in uses of corn come

about gradually and probably take such a Small proportion of the _

ock, of which 40 per cent,

is fed to hogs. 20 per cent to horses and mules, 15 per cent to ~
cattle. 4 per cent to poultry, 1 per cent to sheep and 5 per cent

to livestock not on farms. The remaini

ng 15 per cent is used as

follows: 3.5 per cent for human food on farms, 7 per cent for com-
mercial manufacture (includes that ground in mills). 1.5 per cent

exported, and 3 per cent for other purpo

see. _The proportion used

fer horses and mules has been gradually decreasing, while that for

commerciaquses has been increasing.

, The market selected for the st
is the principal corn market in the Unit
price series, it was necessary to choose
qudtations were available for the entire
sirable to have one in which the volume

udy was Chicago, since this
ed States. In selecting a '
one on which comparable '
period. It was also de-
was as large as possible

and which has been affected least by changes in Board of Trade rules

respecting deliveries on future contract

s. Fer these reasons"No. 3

mixed" was selected. "No; 3 yellow" might have been almost as de-
sirable.‘ An average of daily quotations and of weekly.highs and
lows gave essentially the same monthly averages. '

Since corn is an_annua1 crop,

and distribution of the crop are available, the important supply

factors are known. An average price for some period, of one year
or less,.or for a still shorter period as one month, may be chosen
for study. Eur the longer period, an average of November to May

analyzed by itself finally, but for purp
data for April only are given.

The next step is to determine
price.‘ we naturally select first.those
the greatest effect. The first factor c
general_price level. This was done by
concurrrent Bureau of Labor a11~commodit
have been done by using these as an inde
both methods gays nearly identical resul

the factors which affect this
factors which we believe have
onsidered was changes in the
dividing the prices by the

y index numbers. It could
pendent variable. Trials of .
ts.

aql
1892 was taken as the initial year of the series,

In order to isolate the changes which took place in price
due to changes in annual conditions, it was necessary to eliminate
trend in some manner. It seemed desirable to fit trend lines to
the data of each variable separately rather than use some method of
simultaneous elimination or of correlating differences. Also it
appeared more likely that prices vary as ratios rather than absolutely.
An increase of 300,000,000'bushels in the crop of corn in 1900 would
have a very different effect in cents on the price than an increase
of the same amount in 1928. The percentage changes would be more
nearly comparable. Therefore each variable was taken in terms of
ratios to trend rather than absolute deviations from it. In some
cases a straight line fitted by least squares gave the best fit,
while in other cases the lines were fitted free hand.

Next after the general price level, production was con—
sidered the most important factor influencing the price. The ratios
to trend of the deflated price were correlated with ratios to trend
for production. Prices were estimated directly from the regression
equation from this simple correlation.

The residuals from this correlation were then used as a
new dependent variable and an attempt made to explain them. Stocks
on farms plus visible supply on November l,appeared to have little
independent effect other than increasing the total supply. Accord-
ingly they were added to the production, and production plus carry-
over was considered as one factor. This process was continued and
whenever a factor was found to explain any appreciable part of the
residuals, it was included as an independent variable. Then a
multiple correlation was run and a new set of residuals calculated.

The next factor to be considered was on the demand side,
namely, feeding "requirements". Our data on numbers of various
’classes of livestock, especially previous to 1920, did not appear
accurate enough for this purpose. Only the consumption by hogs
seemed to vary significantly. The data on hog population being
unsatisfactory, some other indicator had to be used. we might have
used market receipts; but since the price of hogs would also do,
and we also wished to take into consideration the price of corn
as a factor in its consumption, we decided to use the corn—hog
ratio as one of our variables and thus care to a considerable
extent for both factors in one variable. Inasmuch as we desired
to make forecasts as early in the fall as possible, the corn-hog
ratio for September calculated from Chicago heavy hog prices and
the price of "No. 3 mixed" corn was employed.

Usually less than 20 per cent of the crop is shipped out
of the County where grown, and less than 10 per cent reaches out
eleven primary markets. The price of the whole crop is based on
the sales of this less than 10 per cent of it. About eleven states
located in the corn belt furnish most of this 10 per cent. These
various states were first considered independtly, and as inter—
correlated: but finally the production in Ohio, Indiana and Illinois

sea

was added together as one factor and that in Iowa considered as an-
other, and that of the remainder of the states was discarded.

Our problem was now in this form: as the dependent vari—
able, the ratios to trend of the deflated price; as the independent
variables, (1), the ratio to normal of the total supply November 1
(production plus carryover); (2), the September corn-hog ratio;

(3), the crop in Iowa; (4), the crop in the three states of Ohio,
Indiana and Illinois. Correlating these factors with the November-
to—May price from 1892 to 1914, assuming a straight-line relationship,
gave a coefficient of multiple correlation of .956. From 1897 to
1927, omitting the years 1915 to 1921 when fluctuations were extreme,
the coefficient was .946. Including the abnormal war years, as a
matter of fact, changed the coefficient very little; but it might
have affected it a great deal. If relationships were the same dur-
ing those years as pre—war and after the war, no harm is done by
'1eaving them out. If the relationships were abnormal and cannot be
'expected to continue at the present time, it is desirable that they
be left out.

There are many other factors that might be expected to
have an affect on the cash price of corn. Some students of the
subject have linked the changes in business actiVity with the prica
of corn. There is a rather close correlation (r =' .78) between the
average November-May corn price and the Harvard ten-commbdity index
'of two years previous. A similar relationship may be found between
corn prices and pig iron, lagged 26 months, and corn and industrial
stocks lagged 30 months. However, the relationship between tota1
production in'stead of price and these factors is almost as great,
while that between the acreage and these factors is small. Unless
one_can explain the connection between these factors and the favorable
or unfavorable corn weather, one is hardly in a position to use them
to explain variations in price. The above mentioned multiple correla~
tion coefficient can be raised from about .94 to .96 by the use of
the Harvard ten-commodity index lagged two years. The correlation
effect between the two may be due in large part to the cycles that
have occurred in both corn prices and in business activity; but in
a period so short as that for which data are available, any coeffic-
ient of correlation between prices and some measure of business
activity includes the relationship between a very'small number of
observations (cycles), with the probable error so great that the
coefficient can hardly be deemed significant.

One would expect to find correlation between the price of
corn, the price of other grains, particul.aily oats, wheat and barley,
and the trading activity in them. Statistical measurem91nt failed to
indicate this to be true for any long period of time once the prices
of these were adjusted for changes in the general price level. The
average November-to-May price is but little affected by other prices
or the amount of speculative activity. For shorter periods of time,
there may be a sympathetic movement which will carry prices along

.393

together for awhile if the fundamental conditions are not too much
against such changes. Even here, price changes may take place for
a short time which will later correct themselves. If a change ought
to take place in a given direction, the movement of prices of other
grains may hasten this. Prices of other coarse grains which may be
substituted for corn become much more of a factor from March or
April throughout the summer than previous to that time.

Likewise, the size of the visible supply and its rate of
increase or decrease are much more important after April 1. Nb
satisfactory method of forecasting visible supply has been worked
out, and hence it is not employed here, since the problem taken
for illustration was forecasting several months in adva e. Fbr
measuring relationships concurrently, visible suppfgfih airly
close connection with price, especially in the Spring and early
summer months. 'Hoever, the visible supply is not always the
causal factor. At times, large visible supply is attended by
high prices, while at others the reverse is true. In the first
case, prices were high enough to bring forth corn at a more rapid
rate than it passed into consuming channels, while in the second
case, the supplies were so great relative to the demand that the
visible supply increased to a high point in the tenninal markets.
As a general rule, a large visible supply in the late spring and
early summer months, decreasing at a rather slow rate, will be
correlated with low prices in those months, and if the probable
visible supply or some indicator of this could be included as a
factor, the probable price in those months could be more correctly
forecastsd.

If,instead of deflated prices, actual prices are used,
the index number employed as a separate variable, and time is used
as a separate variable to take care of the trend by simultaneous
elimination, and each of the factors with the exception of time,
is expressed in terms of logarithms, the index of multiple.correlap
tion becomes .99. From 1892 to 1914, a linear relationship can be
assumed, but if forecasts are to be made for the post war years,
changes in trends that have occurred make curvilinear correlation
necessary. -

Up to the present, time. the forecasts by the two methods
for the postwar years hays been abaut equally successful. The
greatest difficulty in the use of logarithms and simultaneous elimin-
ation of trends is that changes which may be taking place from year
to year in trends and relationships cannot be so easily observed, and
the statistician without a wealth of experience fails to observe ,.
changes which may be significant until long after they have affected
the results. However, once the ratio—to-normal method has been
worked out, the second method, using logarithms and simultaneously
eliminating trends, may aid somewhat. Further checks by the use
of first differences of logarithms, and by absolute first differences,
may give the forecaster further confidence in his forecast, or warn
him in case he cannot see the reasons why the ferecasts by the differ-
ent methods vary appreciably. ‘

do!“

The following constants were obtained for the years 1897
to 1922 (omitting 1915-1921) using the ratio-to—normal method:

R1(2345)

= .9462
31(2345) = 5.11 per cent
, 01 - 15.7870 "
x1. — 255.0188 _ 1.6678X2 .. .338363X3 +.185309X4
"' 024433sz

when:~

X1 8 Ratio to trend of Nov.-May average deflated price
- of No. 3 mixed corn at Chicago.

X3 - Ratio to trend of total supply of corn Nov. 1
(Production plus stocks on farms plus visible.
supply Nov. 1)
.X3 I Ratio to trend of Sept. corn~hog ratio.

X4 I Ratio to trend of corn production in Illindis,
Indiana and Ohio.

‘X5 - Ratio to trend of corn production in Iowa.
Making the proper substitutions, correcting for trend and

prhe.leve1, the following are the estimates and actual prices for
the recent Seyear period for the crop years 1922—27:

Error of Estimate

 

lbv.-May 3 Nov.-May Above or Below

Actual Price Estimated Price Actual Prices
(cents) (cents) (Cents)
1922-23 _ 74.0- 73.7 -O.3
1923-24 76.8 77.9 +'1.1
1924~25 113.5 113.0 -O.5
1925-26 73.8 74.4 1»0.6
1926-27 , 72.5 70.3 ~2.5
1927-28 9131 89.1 -2.0

Pre-war estimates will average about as accurate, the maximum error
being 5 cents.in 1907, when no doubt the formula failed to take into
account the effects of the depression.

495

An average price for a year is much more easily estimated
than one for any short period of time. To forecast the price for
any short period, as for example one month, either f0 two methods
may be reSOrted to. An estimate of an average price for a period
of several months, or a year as above explained can be made, and a
seasonal index applied to this estimate to determine the price for
any singde month. Since certain factors may affect the character
of the seasonal indexes, more than one index will be necessary.

The shape of the seasonal curve for a large crop following a small

one will probably be different from that when a small crOp follows

a large one. It is rather difficult, however, to find enough years

when conditions are similar, so that characteristic seasonal curves

may be calculated. The seasonal curves for individual years seem to indicate
perhaps there is a much less well defined seasonal curve at the

present time than 30 or more years ago. It appears, therefore,that

the use of seasonal curves in forecasting a particular price may be

beset with many of the usual-difficulties encountered in trying to

forecast prices merely with a series of charts.

Multiple correlation analysis of prices by separate
months may prove the more accurate method. The month taken for
illustration is April. It was assumed that prices in this month
are affected by the same factors as have been previously indicated
as affecting the average November-May price. To the deflated aver—
agp.prices of No. 3 mixed corn in April as the dependent variable,
and the other four factors as independent variables, trend curves
were fitted. Trends of prices since the war were taken as horizontal.
Ratios to these trend were employed in calculating_the correlation
coefficient, regression equation and estimates. For the period 1892
to 1914, the multiple correlation coefficient, standard error of es—
timate, standard deviation for the dependent variable, and regression
equation were as follows:

31(2345) = .933
51.2345 3 5.15 per cent
€51 = 16.8348 "

(1) x1 = 241.3000 - .52.0019x2 — .384676x3 —
.035910x4 - .48908915

When the period from 1897 to 1926, omitting years 1915 to
1921 was used the results below were obtained:

R1(2345) I .898
81.2345 = 6.45 per cent
(61. = 14.6659 "

(2) x1 : 233.3044 - 517908Xg - .300883X3 -
.974177x4 - .46243X5

The difference in the two coefficients of multiple correlap
tion resulted in considerable part from the break in prices of all
grains in April, 1925.

.196

The success of any forecasting device can be judged only
by its results. The actual and estimated prices from 1892 to 1928
estimated from relationships existing from 1892 to 1914 as indicated
by the first regression equation, are given in Table I, also the
results using the second regression equation (1897 to 1927). The
production data used here are the final estimates given by the United
States Department of Agriculture in December. Almost as accurate as
estimate can be made by use of November-or-October production esti-
mates. .

A further comparison of accuracy in forecasting can be made
with that of traders in the market, as expresaed in their opinions in
the form of prices of the May future in November or December, and what
these prices actually are in April. '

As previously stated, a drastic movement of futures or
prices of other grains may effect the price in a single month and
add to the forecasting difficulties. Such was the case in April,
1925. The actual price was $1.02. the price estimated by Formula
I, $1.13,and by Formula II, $1.11. The average price for March of
that year was $1.12 and for May $1.11. The price for April should
have been approximately the same. Prices of all grains had reached
a very high level in February of that year. The May corn future Sold
for over $1.38 and remained near this high level until the beginning
of March. Cash corn was at approximately the same price; wheat over
$2.00 per bushel, and other grains correspondingly high. The May
corn future sold as low as 91% cents soon after the first of April,
and all other grains declined from their high in February about a
corresponding amount. The result was as an April price far out of
line with what would normally have been expected. .

The use of an estimating formula is only advocated as an
aid. There are factors that cannot be included in a formula, prin—
cipally because of inadequate data. One of these is the quality of
the crop, particularly its feeding value. The available number of
hogs for feeding soft corn in case Of an early frost and the effect
of movement of large quantities of low grade corn into the visible
supply are further factors. Even after a price estimate has been
made in the fall, subsequent developments may indicate that some
factor used may have been incorrect. Such items as a.larger hog
population than expected, greater disappearanCe between November 1
and March 1. larger carryover on March 1, and a greater visible
supply than expected, are modifying factors. Likewise, the prices
of other coarse grains in the spring, and for a time price levels
of all grains, may be worth considering. No-adequate measures of
the consumption of all classes of livestock, hr the increase of
commercial uses, have been considered here, any appreciable changes
in which should not be neglected. '

297
gable ;, Actual Prices. Estimated. Prices and Errors

of Estimate. 1892 1204928

 

 

/
' :April Price .L; :Errors of :12 :Erro'rs of
Year :No. 3 Mixed :Estimated :Estimate ffmmstiqlated :Estimate from
:, Corn AzApril PricezActual Price :April Pricemctual Price
: (cents) : (cents) : (cents) : (cents) : (cents) Y
: : : : 3 .
1892-93 : 38.9 : 38.0 : -O.9 : :
1893-94 1 37.8 : 38.5 : +0.7 : :
1894-95 : 45. 1 : 43. 8 : ul. 3 : :
1895—96 : 28.8 : 27.7 : «1.1 : :
18961-97 : - 23.0 : 22.8 : - .2 : ":
18976-98 : 31. O z 33. 6 : 7—2. 6 : :
1898-99 : 34.3 : 38.3 : 4.0 : :
1899-1900: 39.5 : 38.6 : -O.9 : :
1900-01 : 43. 1 : 37. 9 ' : e5. 2 :- {W
' 02 a 60.1 : 58.4 : -1.7 : z
03 : 60.6 : 39.8 : -0.8 : :
d4 : 49.5 : 53.5 : +3.9 : z
0 : 47.8 : 51.2 : 43.4 : :
06 : 46.5 z; 48.3 : +1.8\ : :
07 : 44.6 : 46.2 : +1.6 : :
08 : 64. 5 : 58. 8 : .--5. 7 : :
09 : 68.5 t 68.4 : -O.1 : z
10 : 60.0 : 62.2 : +2.2 : :
. 11 : 49.2 : 45.6 : -3.6 : :
. 12 : 76.8 : 71.3 : ~5.5 : :
' 13 x 54.5 : 57.6 : +3.1 2 :
14 1 66.8 : 70.9 : +4.1 : :
1914-315 : 74.5 : 70.1 : ~4.4 - : :
1922-2! : 79. 2 : 75. 1 : «an : 76. 6' : -2. 6
24 : 77.4 : 80.1 : +2.7 : 78.9' x 11.5
25 : 102.0 : 112.9 :3 +10.9 : 111.0 : 9.0
26 : 70.2 : 66.1 : ~4.1 : 67.2 ,: ~3.0
27 : 68.8 : 67.6 : ~1.2 : 70.1 3 +1.3
28 : 101.0 : 97.7 : -3.3 : 96.5 : . -4.5

 

L; Calculated using Fonmila. I based on relationships 189241914
L_2_ Calculated using Formula II based on relationships 1897-1926,
emitting years 1915—1921

293. ~ 3. .nellemggailama.

The method of analogy is at the opposite extreme from
formal statistical method. It bases its conclusions upon a close

,' study of the details of a one or a few cases, whereas formal sta-

tistical method abstracts a few special characteristics from a

large number of cases. Between these two extremes are the case
‘”hhd the informal“statistical methods. The method of analogy is

by Some made identical with the case method, by some called-a vari~
~ant of it, and by others set up as a distinct method. The deci-
sion in this matter depends upon how broadly the case method is
"‘defined on the one hand. and how narrowly the method of analdgy; "
‘The logical difference is that_the case studies_look to generale.‘
izations that can be .applied to a universe, whereas analogy studies
look to statements that can be applied to one particular case, the
one for which the parallel case was'soughti_ To.illustrate, using.
the method of analogy, one would concentrate on.one particular
agricultural depression ~ for example, that in Europe following

the Napoleonic were -.as a basis for forecasting developments in

the present one in the united States. Using the case method, one
would make a complete study of a considerable number of agricultural
depressions as a basis for many generalizations‘concerning such de-
pressions. In the one, the reference is‘always1back to the situations
we are endeavoring to understand; in the other, to uniformities and
differences in the cases being examined.‘ In practice‘ the case study
places emphasis upon the organic unity of each case and the interu
dependence of its various parts a upon seeing the thing as a_whole:
whereas the analogy study places the emphasis upon the answer to a
specific question or two. Then there is the more obvious difference
that a case study may include so many cases that all semblance of an
analOgous approach is_lost. . '

It is not the intent of the framers of this handbook to
inquire closely into the use of analogy as a tool of research in the
field of agricultural economics. The method is indeed widely used -
more widely than is commonly recognized - and no doubt often badly
used. But neither space nor time will permit such a treatment. In-
stead, enough will be said to make us conscious;when we are using it,
so that we may'guard against'its more obwious hazards, For the rest,
the reader is referred to any good textbook on logic.. '

The steps in an analysisiby analogy are as follows:

1. Sufficient preliminary study of the case or situation
at hand so that one can state clearly the elements in
it for which a parallel is sought. .

2. Singling out the parallel cases: This will probably
[require a. reconnaissance study~of a good many of them,

Z?9

Some differences between the two cases may be in-

consequential; others absolutely destructive of all
comparison. The choice must be made with reference
to the elements in the case in hand that have been
chosen for study. ‘

3. Close study of the case at hand and of the parallel
ease to discover likenesses, differences and all
pertinent facts. This will require assembling data
in many cases, from official reports or original
documents, business records, interviews, or perhaps
even from field surveys.

_‘4. Arranging the likeness and differences in parallel
columns so that they can be checked and weighted
against each other. '

:5. Formulation of concluSions, with necessary qualifican
tions. , ,

‘ It is obvious that the method leaves room for a great deal
of personal Judgment. But if it is honestly used and without bias,
this need not be so serious a defect. The important consideration
is that all the pertinent facts, all the likenesses and differences,
be placed in the record, so that those making use of it can decide
for themselves whether the method has been properly-used. Bias enters
by causing certain facts to be rated as more or less important than
they are; dishonesty, by causing them to be consealed.

Analogies may be drawn from many sorts of related fields -
an earlier historical period in the same country or in some other
country, from some other area or country, from some other government,
from Some other industry, some other institution or law, some other
historical incident, Some other person. A student of monetary theory
may draw his parallel from a certain financial-panic; a price fore—
caster, from a certain year in which conditions resembled those pre-
vailing now; a farm management specialist from a certain successful
farm that he knows; a taxation Specialist from a certain state income
tax law that he knows to be successful. .

. The great hazard of the method is that those using it will
fail to appreciate the importance of the differences.- Men supposedly
scholars have been known to argue that finding one case in which
events followed a certain course is sufficient evidence that they
will follow the same course noun There are cases in which this is
true; but not many of them.

The method of analogy has its legitimate uses in research.
The most important of these is in situations where few precedents or
parallel cases are available. The Fairway Farms project in Montana
is worth studying by this method by anyone considering its possible
imitation in another area; similarly the state colonization project
in California; the cooperative egg marketing systems of Missouri or

3&0

Minnesota. National policies with respect to tariffs or surplus
control or land development must be studied by this method if at
all. The second important use is as an early stage of an attack
upon a large comprehensive problem. Just the sort of comparison
that the method of analogy involves, is what is needed. in order to
open widely a problem and reveal its possibilities.

C. The Case Method,"I

The essential na+ ure of the case method is the study rather
much in their entirety of a limited number of units of observation.
Le Play's studies of European families are generally taken as the best
example of case, or monographic, to use the European term, studies.
Bowley says that "In a monograph, after the fashion of 16 Play, a
single family is studied; the occupations and earnings of its members,
the way these earnings are spent, and its economic‘position generally
are set down; but this study is not so far statistical.-— In the
monographic method the individual is everything; in the statistical
method, nothing." Marshall remarks** "that the method of Le Play's
monumental Les Ouvriers Europeens is the intensive study of all the
details of the domestic life of a few carefully chosen families. To
work it well quuires a rare combination of judgment in selecting
cases, and of insight and sympathy in interpreting them. At its best,
it is the best of all; but in ordinary hands it is likely to suggest
more untrustworthy general conclusions, than those obtained by the
extensive method of collecting more rapidly very numerous observations,
reducing them as far as possible to statistical form, and obtaining
broad averages in which inaccuracies and idiosyncrasies may be trusted
to ouuntsract one another to some extent."*** The feature that is
conspicuous in both these statements is the entire absence in case
studies of statiStical summary figures, frequency distributions. dis-
persions, coefficients and the like. Each case stands on its own feet.
The other feature is the emphasis on completeness and unity. The
sociologist, Professor Cooley is even more insistent on this latter.
"Case study,n he says "is a direct and all round stpdy of life histories
as distinguished from more partial and abstract methods" (Cooley is
here speaking of only one type of case study - the life history). An-
other sociologist, Professor Bernard, writes: "The case is not defined
in terms of its simplicity or complexity, but in terms of its organic
unity."

Cooley elaborates further the relation between statistical
method and the case method: nStatistics is an exact method, and it is
enabled to be such precisely because it is not in itself social but
mathematical. It does not directly perceive social facts, or any
other kinds of facts, but it takes standard units of some sort, which
may be perceived social facts, and compiles,arranges, manipulates. and

 

* Prepared with the assistance of Mr. Arnold F. Hinrichs of the
University of Minnesota.

** AA. L. Bowley, Elements of Statistics, 4th edition, 1920.

*** Alfred Marshall, Principles of Economics (8th edition, 1922) p.116.

301'

presents them in a way intended to make them yield ilbmnination.

The statistician operates between the primary observer, on the one
hand, and, on.the other, the theorist who demands light on certain
hypotheses. Evidently it would not be good sense to assume antag—

v onism between the exact methods of statistics and the more fallible
procedure of sympathetic observation and interpretationg‘ They are
complementary and do not or should not overlap. The only opposition
likely to arise is one due to the bias of the practitioner."*

Also Bernard:** "The essential function of case analysis /'
is definition; that of statistics is integration, that is, to look
into perspective and organic unity, in order that the 'mind's eye'
may see social phenomena as wholes rather than as isolated units.
lfihile supplementary, one important distinction between the case and
statistical methods should be noted. A case description is, if
accurate, always a true record of what occurs, while a statistical
‘ generalization, except in those instances when all included cases
are identified, is only an abstract approximation. Definiteness and
concreteness of detail must in.some degree.be sacrificed to the more
inclusive view of the statistical generalization. If frequently
‘cases' or conceptual integrations of phenomena must be broken up
into constituent items in order to look these items into new unities
of perspectives, into new conceptual 'cases,‘ this fact does not
create-an essential antagonism between the two methods. Perhaps it
merely emphasizes their interdependence. The real source of method-
ological error is in supposing that statistical generalization can
proceed with definition of data or units or that generalization can
proceed from case data without some sort of statistical method. So-
called common sense inferences from cases amount to nothing less than
informal statistical generalizations. Formal statistical generaliza—
tions are more adequate, when practicable." ,

. As one would expect, the case idea has been widely applied
since Le Play's time, and a number of different applications have

come to be more or less clearly recognized. The most important

differentiation is that between its use in teaching and in research.

'The value of the cases in instruction was emphasized frtst
in the field of law by John Barnard Byles in England about 1829. He
emphasized "how important it was that the Student should, if he can,
study the law as the judges themselves have declared it, and not as
some unauthvrized interpreter pronounces it to be.”*** It was first

 

* C. H. Cooley, The Boots of Social Knowledge, The American Journal
of Sociology, July 1926, ,L. 73. ‘ .

** "The Development of Methods id Sociolégyf“ The Monist, April 1928,
p. 314. .

*** A.'Z. Reed, Training for the Public Profession of Law,“lfle Carnegie
Foundation for the Advancement of Teaching, Bul, No. 15, 1921,
(History of the Case Method) p. 371.

introduced in the United States by Professor Langdell about 1870 who
is regarded as the reformer of the Harvard Law School.* In_his own
words: "It seemed to me to be possible to take such a branch of the-
law as contracts, for example, and without exceeding comparatively
-moderate limits, to select, classify, and arrange all the cases which
had contributed in any important degree to the growth, development or
establishment of any of its essential doctrines; and that such a work
should not fail to be of material service to all who desired to study
that branch of law systematically and in.its original sources," To-day
the langdell method is used in whole or in part in a majority of the
law schools of the country, ertainly in the best ones. It eventually
came into use along with the laboratory method in medical teaching.

The teaching use which is most significant for our purposes
is in schools of business and courses in economics. It was intro—
duced first in the Harvard School of Business Administration in 1919
’ by Dean W. B. Donham, who considers the "case" to be a step beyond
the "problem" method of teaching, for he says, "I use the word 'case'
rather than problem because the latter fails to connote the actuality
and the realistic detail which must surround the Specific situation
if it is to start with the flavor of life. The case always includes
one or more problems."** ,

The case method has recently been introduced in teaching
fundamental economies to undergraduate students in Columbia College
By W. E. Weld and A. S. Tostlebe. They are joint authors of the
first case book to appear in that field.*** "The method as used in
Columbia involves the case book, a printed syllabus and reference
list, and a text book. The case is used as the basis of instruction
and the student prepares his lesson from the case and reference
material, No written report of the case is required." **** Dis—
cussion is.popular in the classroom. Three criteria are set up,
of satisfactory cases in economics: "First, it must be a story which
contains as nearly as possible an accurate description of some cross—
section of economic life. It should not be faked—-. Second, it

/ ‘must be interesting within itself, with sufficient detail to furnish

1°°a1 C01°rs~-~ Third. it should be significant from the standpoint
of economic theory,fis****

 

* C. C. Langdell, A Selection of Cases on the Law of Contracts;
With References and Citations, prepared for Use as a Textbook
in Harvard Law School, Boston, 1871. '

** Business Teaching by'the Case Systems American Economic Review,
March 1922, p. 57. ' “

_*** -VW. E. Weld and A. S. Tostlebe, A Case Book for Economics, 1926.

**** Panlena Nickell,.The Case Method of Teaching Economics to Home
Economics Students, M. S. Thesis, Columbia, June 1927, p. 33.

***** weld and Tostlebe, P; III. "

308

As much attention as the foregoing is given to the use of
cases in teaching, because along with the tenching of cases has
necessarily gone an extensive program -of research with the objective
of collecting this material. In our early statement of objectives
of agricultural economics research, mush emphasis was placed upon the
need of it as- a basis for better teaching. The early use of the case
method in teaching law would have been'impossible without the centuries
of reported decisions of the courts which form the heritage of common
law. Much of recent research in the legal field has had for its pur-
pose to collect and classify cases. Most-of'this has been done by
the law teachers themselves. This means that it has been done by men
who were familiar with the principles of law, and that the cases were
developed around these principles. The case method therefore became
another and better way of teaching the recognized principles - not
these principles in their old phraseologies or even in their exact
content,.but in forms more realistic and vital; It introduCed a ,
more objective and inductive, and hence more scientific, approach ///
to the study of principles both in teaching and in the supporting
research. . .

The extension of the case method of teaching into other
fields requires the creation of effective substitutes for the re—
corded cases available'in law. Dean Donham recognized this at the
start. The Harvard Bureau of Business research collects large nump
bers of cases. according to two different methods: First, the in-
structor may go directly to a business man knowing exactly what he
wants and obtain his exact data. -Reality is injected into such
cases because they come from the executive's own experience. The
principal disadvantage of this approach is that the large amount of
travelling and interviewing requires too much of the instructor's
time. A second and better plan, because of less expense involved,
is to employ field agents, Harvard graduates,_to act as Collectors
gathering the data guided by a fully developed outline. Five persons
supervise the work of some thirty field agents. and examine and class—
ify the materials obtained. The agents travel in all parts of the
country to obtain a diversified collection of cases. Professors
decide for which particular fields cases shdll be collected. Their
wishes are made known through the superviSors to the proper field
agents who proceed to locate and obtain such cases from business '
firms. The agent assembles a case and sends it to his supervisor
who sends one copy to the professor, andther to the ftnnfrou Which
it was obtained, and files a third in the Bureau. The instructor
detennines the usefulness of a case for academic instruction and
is responsible for the final form in which it is published. The
aim is to obtain cases met most frequently in,the. business world
rather than those depitting unusual circumstances. Since business
heads and directors frequently do not recognize the existence of
such cases in their particular business, the field agents are spec-
ially- trained to detect them. .

The Bureau employs two additional men charged with pub— .
lishing the Harvard Business Beports.‘ the first.s 550 page volume
was published in 1926, and contains 149 out of 3500 cases collected.

at .
* Harvard Business Reports, A. a. fiaw, Ehicago, 1§§6.

Two volumes have since been published.

In addition to these reports are a dozen or more case
books arranged for specialized study in the various courses.**

The cases published in the Columbia case book were col-
lected from a variety of sources, including (1) a small amount of
personal investigations, (2) students in the course who are re-
quired to hand in two cases a year as part of the instruction
requirement, (3) suggestions from publications. and (4) the Harv-
ard Bureau of Business Research.

Obviously business and economics have been at a great
disadvantage as compared with law in the use of the case method.
Not having the cases on record has been by no means the greatest
handicap. It has been instead that the business cases do not re-
veal so clearly the principles inVolved as do the legal cases. In
fact, the report of the legal case usually includes a specific re-
ference to the principles involved.’ The law cases have been taught
by men well informed as to the principles. If they did not know at
the start the classroom technique for development.of principles from
cases, they have had a long time since to learn it. The teachers of
business cases have tried to develop principles and cases and teaché
ing technique all at the same time. They could.have helped them—
selves greatly if they had been willing to recognize that the prins
ciples upon which business management is founded are largely identic-
al with those of economics, But for one reason or another many of
them have either refused to accept this; or have felt that a wholly
fresh approach is desirable even if it finally leads them to very
nearly the same body of principles in the end.

No doubt economics itself is largely responsible for this
attitude of the case—makers. The economic principles stressed hays
been more largely those of social and political economy than of
private economy. The applications have largely been to public prob—
lems. Even then, they have been left in the form of broad general
abstractions, not reduced to a "working basis",as the makers of cases
must have them. When economics gets down_to the details of private
business affairs, and begins'to state its principles in terms of them,
the teachers of business will begin to build their cases around econ—
omic principles Just as the teachers of law have built theirs around
legal principles. A rather wide adoption of the case method by
teachers of pconomics will do much to convert the principles of econ-
omic science to the form desired. Just as the case method has made
the principles of law more realistic and vital, so will it make the

yrinciplegpf economics. A p c
* Among the earliest of these backs may Eta—mentioned the renewing:
a M.T. Copeland, Marketing Problems, 1920;
b D.[. David, Retail Store Management Problems,'1922.
° 1.5. Dewing, Problems to accompany the Fincancial Policy of

Corporations. 1921.
3:; E.E. Lincoln, Problems in Business Finance, 1921.
e

 

I. Tosdal, Problems in Sales Management, 1921.
. B. Vanderblue, Problems in Business Economics, 1924.

‘9‘

p 305
Following are some of the other differentiations that“hawe‘ ““““ ~—~
occured as to the case idea: '

1; As to choice of cases-

(a) “Type" cases - such as several different types of
dairy farms. .

(b) "Typical" or "normal" cases - a few cases each
approximating the model or usual as nearly
as possible..

(c) A normal frequency of cases.

. .(d) Abnormal types - as in social pathology work.
Or perhaps a highly successful farm.

2. As to number of cases-
(a) One only. L /

, (b) Several, sometimes nearly alike, sometimes partly
alike and partly different.

(c) A considerable group.

3. As to types of units of observation - individuals, groups,
communities, areas, institutions, business firms, etc.

4. As to completeness of analysis~
(a) Including all aspects and details,

(b) Including only one or a few specific aspects or
problems.

5. As to research objectivesé

(a) To obtain detailed information of actual operations,
methods and practices - as one might study
closely the activities of a particular chain-
store system in order to learn how it operates.
The effort may be limited to a single opera~
tion, such as hedging by grain elevators.

(b) To get the exact data of some unit such as a farm so
as to have them as a basis for figuring out
trial reorganizations and the like; or testing
out different systems of analysis or bookkeep-
ing.

306
- (c) To explore thoroly some unit in order to see
it whole and sense all the interrelations;
or to get new ideas as to problems to in-
vestigate, improve perspective, broaden
outlook.

(d) To get accurate details as to same one or a few
problems from several units nearly alike ~
as one might study labor conflicts on sev-
eral very similar corn belt farms. If a
larger group is included, theywnmy be
chosen so as to represent several sets
of conditions.

(6) To obtain comparative data from several units more
or lees different in the respects in which
one is interested. Thus.one might make a
comparative study of three large chain-store
systems or line-elevator systems, or of
several farms with different proportions
of certain crops in the rotation systems.

Our interest in the case method at this point is as a
method of analysis. The methods of collecting the data of case
material were discuseed in an earlier section. Closely supervised
records will generally be needed where the farm is the;unit of obser-
vation. Business firms will frequently have kept records in suffi-
cient detail.

I

r .

.The method cf analysis must of course vary with the research
objectives. Let us first take as an objective the seeing of the unit
as a whole with interrelated parts. 'The undertaking here is one of
qualitative analytical description, to which a special section will
be devoted later. The.essentia1 features of it are setting down in
visualizable form the various parts ofrthe structure being examined,
or operations involved in its functioning, and speculating as to why
they fit together or work together exactly as they do; then setting
up a group of hypotheses about this; then going back to the sources
of'information all over again and observing more closely having in
mind these particular hypotheses; setting up new hypotheses if need
be: and so proceeding until a rational theory is built up as to why
things are as they are. If the unit of observation is an area, map-'
ping will be found a powerful aid to the mind in such analysts.

‘ It is recognized that further evidence_than this may be
needed n from other cases, or from a statistical exanineticn of a
large number of them — before any generalization can be made of
wider application than to this particular case: and even before
any statements can be made with assurance with respect to this
.particular caseI since ordinarily it must be demonstrated that
the theory holds for most other cases as well before it is accepted
for any one; or ample reasons discovered as to why this particular

807
case in an exception to a different theory. Oftentimes the theory
may be one commonly accepted in its general form, and what is wanted
is an understanding of variant details of it. The qualities of mind
required for such analysis are imagination and ingenuity. It re-
sembles a game of puzzle pictures in certain respects.

If the objective is merely to obtain the facts as to prac-
tice and procedure, no setting up of hypotheses is involved. All
that is necesaary is to check the facts to observe that none is '
missing or is incorrectly stated. A fairly careful preliminary re-
port prepared on the spot will be found a useful device to this end.

If what is wanted is data that can be used in sample an—

‘ alyses, or in working up typical reogranizations, one must have

definitely in mind in advance the manner in which the data are to
be used, the statistical units needed, the qualitative of
the problem, etc. Ordinarily.a definite schedule should be pre—
pared in advance.

Nothing has been said thus far about the choice of the cases.
The earlier discussion of sampling‘has a limited application here. If
the cases are intended to be typical, or to constitute a typical fre-
quency, oftentimes a preliminary survey of the area or group will be
necessary in order to locate the modal or characteristic frequency
units. In other cases, sufficient information for this will be al—
ready available in official records. It must be remembered that one
has no right to present information as that of a typical case unless
one demonstrates that it is typical. If a number of different type
cases are to be examined, it is equally important to determine in
advance that each represents important groups of existing types.

In case studies clearly of the comparative sort, the first
problem is to state exactly the differences in structure and function-
ing, and practices and procedure; and the second to relate these to
differences in the results obtained. The second of theSe calls for
the sort of speculating and testing of hypotheses that has already
been described.

The foregoing types of case studies are likely to be used
most frequently in the early stages of the development of a field -
in its exploratory stages. Presently a point is reached where the
need is more for detailed examination of certain problems or rela~

'tionships. The case study then, instead of being aimed at includ-

ing everything in order to see the whole, is directed at answering
precise questions as to certain facts of it. In a study of grain
elevators, it may be aimed.at hedging and storage operatidns, and
buying policy. In a case study of a group of farms, it may be
aimed at the effects of keeping different numbers of cows, of feed-
ing dairy cows at different rates, of having them freshen at differ—
ent seasons, and a number of other matters relating to the dairy
enterprise, all of which require detailed information. It may be
aimed at answering a group of questions relating to conflicts of

80'

labor and to crop combinations, Where Inch is the case, the prob»
lems or questions must be carefully stated in advance and the method
of analysis determined. A method of attack suited to some questions
is to_find a smallgroup of units which are modal or nearly modal in
relation to the problem and work up separate results for each, note
the differences and likenesses, and from this analysis select what
seems most nearly representative. A statistical average will seldom

and mode from it. The analysis in.this case verges on the statist&3--
ical.‘ If more units are included, they may be selected so as to
fall into two or three groups, each to be worked up separately as
outlined above, so that comparisons can be made between the groups.

set of conditions needing to be studied and presenting the results
’ by individual farms or small homogeneous groups of them. As a
device for presenting the results of such'analysis, a table of
figures for good and bad farms'may be highly serviceable; but it

. One type of caSe research is sometimes called the "mul-
'tiple monograph" method. This term is applied when a number of in-
dividual studies are made usually more or less according to one out-
line of work. Oftentimes the different monographs are prepared by
different individuals., "Even if one person makes all the studies,
he is sure to find his ideas changing as he goes forwards Hence
there still remains at the end of task of coordinating the separate

early stage of development of a field of research. Even having
different persons work up the different studies has its important

Life History of an Immigrant" by T omas and Znaniecki, 1919. Fbster
and Anderson's one hundred brief case histories describing behavior
_.of:cht1dren between the ages of ten and six, published in 1927 as

’309

"The Young and His Parents" is another example. Professor S. B.
Gra's study of the English village of Crawley must be mentioned

as an example in the historical fieldr Many case studies have

been written of labor unions. Laur makes use of case data of two
Swiss farms, Nehhfif and Sentenhof, in presenting his plan of an—
alysis of the farm business. It is clear from the early statements
of objectives of the farm management route studies that the sponsors
of these had in mind the use of the results as case material prim-
arily. It was then felt that teachers of farm management knew too
little about ibtual farming and its problems, and that only detailed
study of the case type would provide the information needed. Pres—
ently, however, when the research point of view came to the fore in
these studies, a specific method of analysis, partly accounting and
partly statistical, was developed deductively - also borrowed to a
certain extent from the industrial field — and substituted for a
"case type of analysis. Statistical averages of various sorts were
computed and used in place of the data of individual farms. The
policy followed in the more recent case studies in the business
field has been significantly different from the foregoing. In
place of imposing a method of analysis of their own, these case—
makers have endeavored to find out inducttvégy how business execu—
utives have themselves analyzed their problems, and it is these
methods that they have carried back to the classroom and even used
in their own analysis of similar problems. We do not even yet,

in spite of years of collecting of case data, know a great deal
about how our successful farmers analyze their own problems.

It is equally clear from Mr. Hutson's description of pre—
sent route studies* that most of them are now coming back to the
original case idea, although with the important difference that the
attention is new bling flinhered increasingly on certain specific
questions, such as estimating the effect on net income of varying.
the crop and livestock combinations, varying the amount of the out-
put and size of business, etc. On the other hand, some route studies
are taking in more farms and going over to a statistical basis. More
effort is also being given to finding out how the successful farmers
have solved their own problems. The methods of analysis now being
used are increasingly like those used by successful farmers in so
far as we know what these are. ‘

D. Informal Statistical Method.

Obviously what is here designated as the informal statist-
ical method, even more than the case method, is difficult to confine
wihhin the boundaries of a definition., It may differ from statistics
in several ways, (a) in that the cases are too few to make frequency
distributions, averages, coefficient, etc. of much significance; (b)
the cases are so unlike that it is impossible to abstract common at-
tributes from them; (c) the cases are both too few and too unlike;

 

* See section on "Supervised Records" in Volume I.

;310

(d).thc attributes common to them are not measured at all, but only
roughly eValuated, or perhaps only their frequencies are counted or
estimated; (e) insufficient evidence is available as to the sampling-
often only special or abnormal cases are available; (f) the whole
process of abstracting and approximating is done unconsciously.

, It is impertant to distinguish between the conscious and

the unconScious type of informhl statistics. The former is used by
intent when the Cases are too few or too alike, or when the attributes
cannot be-measurcd, or the sampling is uncertain; the latter not only
under these conditions, but often when the data are adequate and meas-
urement entirely possible, but formal statistical analysis is avoided
for various reasons. .Thus if a grain inspector were to tell you that
the test weight of wheat figures little in the price, he would prob-
ably be-employing, unconsciously, the method of informal statistics,
Formal analysis might easily reveal that unconsciously the inspector
had selected only abnormal caSes as a basis for his conclusion. a

The memory sieve has a way of letting the usual slip between its
meshes. ‘ ' ' "‘

, - 'InfOrmaI statistical method differs from the case method
in that its data are approximate and general, when those of case ,
studies'are exact and detailed. So far as generalizations are made
from cases. it is by the processes of informal statistics.

. The excerpt given on Page 16 from Professor Bernard's ar~
ticle should be re—read at this point, together wit“ the following
additional excerpts:* '"The analogical and the informal statistical
methods contended for mastery in the minds of early thinkers, as
today they eontcnd in the mind of the man in_the street. Nor are
the survivals of these methods unknown among’the sociological ex-
perts. "Other early examples of quantitative generalizations_form—
ulated by the informal statistical method are to be found surviving
- in ancient literatures and folk traditions, in proverbs and popular
wisdom,.in Weather lore and in the survivals of magic and supersti—
tion; Aristotle, who wrote the first formal treaties on society,
made repeated use of this methbd, as well as of the more deductive
method of analogy. ‘ '

The philosophers of history also employed the informal
statistical method along with the case method and analogical deduc-
tions. With the growth of historical perspective and the accumula—
tion of historical data to the point where historical facts began
to repeat themselves approximately, the informal statistical method
could be used in constructing historical sequences, and thus it was
largely employed. Many of the numerous theories of social evolu-
rtion, of stages, and of progress were arrived at,,at least in part,

' i

*‘ L.'Lr Bernard, Ehe Development of Methods in Sociology, The
Monist, April 1928; p. 311-313. ‘

‘, 311

through the application of this method. But this method was always
used primarily in generalizing contemporaneous data. It was particu—
larly useful in the early national and regional surveys, where also
the formal statistical method began to be used. The writings of such
men as Machiavelli, Bodin, Montesquieu, Turgot, Herder, Rousseau,
Saint—Simon, Godwin, Comte, and Buckle, abound in informal statistical
generalization about various sorts of contemporaneous and_historica1
items, as do the writings of present-day publicists.

The development of formal statistics as a method of sociolog—
ical generalization is primarily the product of the nineteenth century
applications of quantitative measurements to social phenomena. It was
made possible by the collection of vast quantities of data of the same
class or order in fiairly restricted fields, since it is not possible to
| apply the method except to collections of data which provide fairly
easily and adequately representative sample. Thus formal statistics
was first applied to population and financial and industrial data, to
the generalization of accurate formulas and principles regarding death,
birth and marriage frequencsies, and had been extended increasingly to
such subjects as circulation of money and credit, exchange volumes and
rates, wholesale and retail prices, business cycles, price prediction,
and the like. Although the fOrmal statistical method is obviously much
more accurate and dependable than the informal method, it is not pos—
sible to substitute the former for the latter in all cases, at least not
immediately. The formal method works best with large masses of data,
and especially must these data be comparable or of the same order. In
many types of social situations the approximate recurrence of phenomena
under relatively the same conditions and within a limited period of
time is so rare than even informal statistical generalizations are
made with difficulty. In such cases analogical deduztion continues
to be used, or, in the event of the failure of this method, each case
or item is regarded as unique and no attempt is made to construct gen—
eral principles, laws, or formulas."

It is obvious that as the use of the informal statistical
method becomes more conscious, the dangers attendant upon its use
rapidly diminish. It tends to smorge under these conditions with
elementary types of tormal statistics. The method of analyzing
prices practiced by Professor Green of Kansas may be classed either
as informal statistics or elementary formal statistics. Professor
Green's description of it is included at this point because it il-
lustrates what happens when the processes of informal statistics
are consciously performed. '

Informal Price Analysis*

A starting point in the analysis may be the determination
of whether the present price level of the commodity under considera—
tion iS'high or low. The index number and purchasing power figures
are the principal data used for this. If, relative to the price of
other commodities, the price of the particular commodity is low, the
presumption is that it will advance and vice versa. Proceding upon

*By ProfessorR. M. Green, Kansas Agricultural College.

 

313
these lines the next step is to measure the purchasing power available
to the buyers of the particular product under consideration. This
involves studies of production in the fields engaged in by a large
part of the buyers of the particular commodity. It involves further
the study of their local money and credit conditions, interest rates,
and general trend of prices in buying territory. Next, purchasing
power and credit in total that is available for all industries is to
be studied. In considering this aspect of demand the agricultural
eccnomist generally accepts the results of other current studies
rather than go so far afield in his investigations.

‘ I

Having considered the principal measurable elements in demand.
attention is given to supply. In the case of some staple agricultural
.products. such as wheat, supply changes will be far the more important
relative to price changes. ‘This is so to a lesser degree in the case
of hogs. and no doubt still less so in the case of potatoes and milk.
It is only reasonable, therefore, that methods must vary especially as
to emphasis laid upon supply or demand or upon certain aspects of each.

In considering supply influences, a first step is the determina»
tion of whether estimated total supplies are large or small. This
necesSitates the use of relatives of some kind. Moving averages, de—
parture from secular trend and such satistical devices are the ones
used. A large crop is taken to mean lower prices and a small crop,
higher prices.

The character and time of the-ihiftingprice levels will,
however, be influenced by the sequence of different size crop yields
as well as by the difference in yields. ' For instance. a small
crop running into anothe'small crop is likely to have a different
shaped seasonal price movement from that of a small corn crop run—
ning into a large corn crop, and so with other sequences. The last
half of crop—year prices in particular is affected by the seQuence.

Having arrived at some Judgment as to whether prices are
likely to move to higher or lower levels and what the time and
character of the movement are likely to be, based on size of crops
and sequence, other aspects of supply are to be considered.

The next step in the procedure may be a determination of
whether the relation of recent prices, reflecting most recent causal
influences, to earlier prices, reflecting more distant causal in--
fluences, is such at some "strategic" point in hhe year as to
indicate that prices will tend up or“ own. The purpose served is
further confirmation of direction of trend and a nearer approximation
to the time of its beginning. Scatter iagrams, graphs of prices
for individual years and frequency tablgp are statistical devices
that may be used.

The next step, after general direction of price trend has
been Judged, may be tordetermine whether there are any times in
the crop year when supply and demand influences frequently become

so closely equated that the scatter or average of prices at those
times is close to the season‘s average price. This gives a pro—
jected season's average price after the lapse of some months that
is useful as a basis in judging whether the current price is near—
ing a high or a low point. '

The next step is a determination of the frequency with
which prices at any season advance or decline and an association
of the probable causes with the tendency. Frequency tables,
simple correlations and individual graphing are useful statistical
devices. Frequency of cccurence in this instance is not used to
indicate probability in the sense it is used in mathematical theory.
This is the case because the data referred to are not random samples
and are considered only with reference to recurrent causal influences
known to operate at the season of the year under consideration.

The need for studies of shorter—time price fluctuations
can only be pointed out. Methods that show any consistency in re—
sults or permit of reasonable inferences are yet to be worked out.
An example of wheat prices will illustrate the problem involved.
Average daily prices for top No.2 hard winter wheat at Kansas
City in January and February, 1923, were $1.20 per bushel both
months. Top January price was $1.23 a bushel and top February
price $1.2M a bushel. At one time in January, the price was up
as much as 6 cents in six days and then at another period down 6
cents in four days. Likewise in February, the price was up 7 cents
in one 10—day period and in another nine days was down 8 cents. Such
price movements are not small enough to be ignored, although they
can be smoothed out in averages. It is just this situation that
makes any average or carefully weighted format statistical measure
of central tendency more significant when the purpose is to approx-
imate a price level existent over a considerable time and which is
the result of numerous up—andadOWn movements that tend to cancel
each other with reference to the longer time average price level.

For these reasons it is believed by some that the Pearson—
ian type of analysis is most useful where the problem is one of ad. “
justing production with reference to price-levels likely to prevail
over a considerable period of time in the future. It is less well
adapted where the problems are planning when to sell, the length
of feeding period, type of feeders to buy or similar marketing price
problems that arise after crops and livestock have already been pro-
duced or bought.

As nearer—atvhand price fluctuations are considered, the
necessity of considering other aspects of supply than total supply
and movement of supply becomes more urgent. Quality or composition
of supply is frequently an important factor in nearby price movements
and may even affect the results of longer time forecasts by the form-
ula method. As an example, low No. 2 hard winter wheat and top No. 2
hard winter wheat, do not always follow parallel courses. The first
is essentially an export type of wheat, while for the second there

3L$

is usually good mill demand. An average of daily high and low No. 2
hard winter wheat prices, therefore, is‘a purely hypothetical price
not closely applicable to either of-these two market types of wheat.

Likewiée, IOCation or position of supply with reference to
censuming industries affects current price movements’ Thus wheat
prices probably follow Livcapool prices with different degrees of
closeness according to the 8138 of the Kansas wheat crop relative
to the size of the world crcp. ' ‘

Such observations emphasize the importance of having any
statistical analysis in a form that will readily admit the use of
new material whenever nearby price changes are being analyzed. Be~
Cruse scope of analysis is even more important than preciseness of
measurement, scatter diagrams, frequencies, individual graphing and
other devices that do not obscure details are useful tools in arriv—
ing at price Judgments.

E1.” The Experimental Method.

In view of the increasing interest in so-called economic
experiments} the committee thought it well to ascertain the state
of the thinking of agricultural economists on the subject, and sent
out the following letter to all departments in this field in agri~
cultural colleges:- ~ a

"We realize that you have already been troubled enough
by the committee on the research-handbook, but there is one more
subject upon which your help is needed, namely the possibilities
and limitations of the use of the experimental method in agricul—
tural economics research. t is a subject to which economists
have thuenily returned with renewed interest. Professor Mitchell
discussed it in his presidential address before the American Econ~
omic Association in 1925.

_ "Experimental method in the recent discussidn is no longer
being taken in the rigorous sense cf_holding all the forces constant
except the one in question; but rather in the sonse'of holding part
of them fairly censtant and measuring the rest of them. As a matter
of fact, a great deal of biological experimentation fits such a des~
cription. 'The difference is only a matter of devree.

"The subject of the experimental method in agriCultural econ~
omic research is being considered in the following connections:

1. Farm organizations closely resembling

types desired to be studied are singled

out and observed closely for a series of
years, adequate records being kept of all
matters bearing on the results under - o
‘question. " ' '

a . . /

3£5

2. In addition to the above, changes of the
sort desired to be studied are introduced
in cooperation with the operator and the
results and observed.

3. Farm organizations of the desired types
are actually set up and observed in
operation over a series of years; changes
may be introduced from year to year.

h. Same procedure as l, 2, 3 for marketing
business units, farmers' clubs, leasing
systems, etc.

5. Technical matters of especial economic
'interest are tried out experimentally;
e.g., sales methods, advertising media,
grading of grain by different graders,
consumer's preference, etc.

6. Technical results needed by economists
are worked out in cooperation with
technical departments: e.g., diminishing
returns in use of fertilizer or feed.

"we would like you to tell us briefly of any instances
you know of that come under the above or any other heads, and where
we can get additional information concerning the methods and results
obtained. Also what are your reactions as to the value, pessibilities
and limitations of work along these lines."

Of the 30 departments which discussed the subject, 20 Were
of the opinion that some piece of work which had been done in their
state, or was going forward or was being planned, could properly be
described as experimental in some degree, under the very liberal
definition given in the inquiry.- The definition was purposely made
broad so as to stimulate seme kind of a.response to the-inquiry. In
this respect, it was more than successful, for a number of departments
were apparently suddenly made to feel that they had been using the ex~
perimental method right along without realizing it. Others, however,
were disposed to adhere to a more rigorous definition and ruled out
similar projects at their stations.

The discussion can be opened in no better way than by
reading carefully ProfeSsor Mitchell's suggestive discussion of it;*

"In speaking of experimentation, I do not forget the difficulty of
making experiments in the social sciences. ~That difficulty seems
to me almost insuperable, seglong as we hold to the old conceptions

 

 

* American Economic Review, March 1925, p.8~9.

. 9‘7“;

labored. For we can try experiments upon group behavior. Indeed,
we are already trying such experiments. We have experimental schools,
in which the physical and social environments of the children are

stimuli offered by the schools and the learning response. So, too,

we experiment with different systems of remunerating labor, different
fonns of publicity, different organizations for distributing products,
different price policies, different methods of supervising public
utilities, and the like.

"Of course, these experiments upon group behavior lack the
rigor of the experimenting done in physical laboratories. The limits
within which human beings can be manipulated are narrow; the behavior

precautions. The ideal of a single crucial experiment cannot be fol-
lowed. The experiments must be-repeated upon numerous individuals or
groups; the varieties of reactions to the stimuli must be recorded and
analyzed; the representative character of the samples must be known
before generalizations can be established. This whole procedure may
have more in common with the quantitative study of data drawn fran
common experience than with the procedure of the man who deals with
electric currents passing through a vacuum tube.v But thatever ap—
preaches are made toward controlling the conditions under which groups
,act will be eagerly seized upon and developed with results which we
cannot yet foresee.

"In collecting and analyzing such experimental data as they
‘can obtain, the most exacting_opportunities for developing statistical
technique — opportunities even finer than are offered by the recurrent f
phenomena of business cycles. It is conceivable that the tentative
experimenting of the present may develop into the most absorbing activ-
ity of economists in the future.' If that does happen, the reflex in-
fluence upon economic theory will be more radical than any we can expect

from the quantitative analysis of ordinary behavior records, The most

4_\/

cussion are: First, that combination with the statistical in the way

of measuring the factors not under control and isolating these effects
by correlation analysis, and also testing out the rspresentativeness

of the cases under experiment,.may achieve some of the same results as
holding all conditions under control; Second,-that the same ends may at

. times be achieved by performing the trial with a considerable number of
individuals - in this way the personal_fagt9rmmay be isolated, or perhaps

317

balanced out in some sort of an averaging process. Repeating the trial
in different years would work in the same way with year—to-year varia—
tions in Weather, prices, etc. Third, that by operating with groups
rather than individuals, the personal factor can be eliminated. We
may not be able to forecast the behavior of any one individual; but

we can forecast how many in a hundred will react in a given way. It
should be added that this last suggestion is of little assistance in
experiments in which the individual farm or businesS firm is the unit
of observation.

There are those also who will rule out most of what Professor
Mitchell has in mind as not experiments at all in the accepted sense
of the word, because not set up solely for the purpose of observing
certain effects regardless of other consequences that may ensue. The
set—upsProfessor Mitchell is talking about mestly — for example, differ-
ent ways of remnnerating labor — are expected to prove successful or
they would never be tried - failure would be too costly. If one ac—
cepts such things as Professor Mitchell's lists as experiments, then
every trial of a new method or practice, every new venture, is in
danger of being called an experiment; and perhaps even old methods
and practices every time they are repeated.

While probably the foregoing is overstating a position, we
will all agree that merely "trying out" something does not constitute
experimentation. Neither dBGS that method of making progress which
is commonly described as "trial and error." If a trial is to qualify
as an experiment, the set—up must be special to some extent — with
some of the usual disturbing factors removed; and there must also be
more than the usual chance that it will turn out badly ~0therwise it
will have to be called a "demonstration" and not an experiment.

It is clear that a majority of those who discussed the prob—
lem were thinking of demonstrations much more than they were of ex—
periments. They talked of the difficulties of discovering with suffic-
ient certainty in advance what will succeed, or of the difficulty of
convincing farmers in advance that the proposed change will pay, etc.
Thinking along these lines, they then pointed out the commonly recognized
faults of demonstration fanns, namely, that they do not show what can
be done under actual conditions, but under specially favorable conditions,
with highly skilled management or supervision, etc.

Let us new take a group of farms such as most of the corres-
pondents had in mind and see if we can set them up=in a form that will
really qualify as an experiment. We will assume that conditions in the
area are fairly well known, but that there are different theses held as
to some feature of the farming practiced there 7 some favoring, let us
say, much cotton and a little livestock and feed crops, and others con—
siderably less cotton and more of the other. As a first step in setting
up the experiment, no doubt a survey wOuld need to be made of perhaps
two hundred fanns, first, in order to obtain a fuller knowledge of all
the variations in conditions, and second, as a basis for selecting the

twenty farms, let us say, to be used in the experiment, Next would
come_the formulating of the terms of the experiment — deciding what
conditions would be as nearly the same on all farms as practicable —

e. g., size of farms, soil type, tenure, ability of the farmer, family
labor, quality of livestock, machine equipment; second, which would be
allowed to be different, but measured as statistical variables — ob~
viously weather would belong in this group, and size of farm might be;
and third, which one would be-varied for the sake of noting the effect

of it‘— namely, crop and livestock combinations in this case. But
obviously not all possible crop and ltvestock combinations could be
considered — a selection would have to be made, perhaps of 5 different
combinations and h farms each, these representing different degrees of
combinations; perhaps of as few as two different combinations, lO farms
for each. Next, the farms would be seledted. This would be a difficult
task indeed. The ideal would be to have the farms in each set as nearly
alike as possible except in the one respect. Farmers would have to be
found who were willing to comply with the needs of the experiment. There
would be the danger that only those accustomed to a given combination, or
especially situated for it,.would be-willing to try it. Those trying out
new combinations would not learn to handle them at once. When the BX“
periment was set up in operation, there would need to be acme helpful
form of supervision, and also provision for keeping certain records,
preferrably not the complete records of the usual route studies, but in-
stead a special set of records calling only for the information needed,
in the first place, as a measure of varying conditions on the different
farms, and second, as a basis for calculating, by the simplest and most
direct method, the results of each combination. Some approximate record
would probably be needed that would indicate how hard the different farm
families Worked. Four farms would probably not be enough of a group to
give the effect of differences in management and the like to compensate
in an average. Hence the one experiment would not be conclusive. Two
groups of ten each would work better in this respect but throw no light
on the intermediate combinations. Moreover, either some way would need
to be found for adjusting for the special conditions of yield, prices,
etc,, prevailing during the year; or the experiment would need to be
continued for several years. It would probably need to be continued
anyway, because many of the effects of the different combinations ~

such as those on soil, disease and weed control — take time to appear.
Herds of livestock need to be developed before they are very profitable.
Given enough different farms over a long enough period and enough differ-
ent combinations, it could presently be demonstrated what combinations
paid best under different conditions. The method would surely be'lab—
orious and costly for those conducting it; and many farmers would lose
money trying out what proved to be less successful cembinations.

As contrasted with experimentations in the laboratory or ex~
periment station feedlot, it would have the important advantage of being
-carried an under conditions more nearly like those on actual farms, and
of giving results more nearly applicable in real farming. ”

519

As contrasted with statistical method applied to a somewhat
larger number of farms in the same area, it would have the advantage
of more precision, provided enough-farms were included in-the ex—
periment. As between a small group handled experimentally and a
considerably larger handled statistically, the advantage could easily
lie with the statistical attack; and the statistical attack would be
the less expensive.

If, however, not enough actual farms can be found which
are practicing a system of farming that needs to be tried out, then
special farms need to be set up, which means some sort of experi—
mental method. This Was the situation in Montana. At the start,
no farms could be found in that area which were practicing the
system which Mr. Wilson wished to have tried. '

Needless to state, no experiments meeting exactly the
above description Was reported from anywhere. The Fairway Farm
project in Montana Was started with the purpose of trying out
certain ideas as to scale of operations, use of machinery, crop
and livestock combinations and tenure, which it was believed in
adVance would prove the solution of the farming troubles in the
area. These ideas have been adhered to in the main since. But
the project as at present conducted is most accurately described
in the first instance as an offibrt' of skilled overhead management to
to make each one of a group of farms“ successful, and to work out
the plan of organization suited to the farmer and the available
resources that will be most profitable. To use Mr. Tilson's own
language: "As a matter of fact, each of the FairWay farms had
boiled itself down pretty much into a. concrete example of a.
hypothetical farm worked up to make the best use of the re-
sources, land and managerial ability involved in each specific
instance." This means, on one farm, renting some additional
land; on another, getting a larger tractor; on another, growing
some more forage; on another, milkingsome more cows; on another,
more brood sows in place of buying feeder pigs. If a farmer
twists his knee in an accident andhelp cannot be hired at
planting, it means letting the crop land lie idle that year.

The most important reapect in which the project is ex-
perimental is that certain eseential ideas as to farm organize;
tion concerning which there was some uncertainty, are being tested.
Furthermore, each of the new departures undertaken on each farm
has somewhat more than the usual amount of uncertainty as to its
outcome;. All in all, it must be Said that new departures are
being undertaken more rapidly and with a bolder freer hand than
ordinary conditions would warrant or the ordinary farmer would
venture. This is made possible by- the control of conditions
which the Fairway Farms financial arrangements makes possible.

sea

,I From the standpoint of technique of experimentation in farming,
the Fairway Fame has this of importance to offer - the changes intro-
duced are not continued long enougi to test them thoroughly, as above
outlined, unless they turn out profitably or give promise of it early.
Instead, Mr. Wilson presumes to be able to allow for abnormal condi— _
tions - of which Montana offers an abundance — and prognosticate what
would happen in the long run, after a rather limited trial of a new
departure. To the extent that this can be done, and, done with reason-
able certainty, it will save a great deal of time and expense. Records
suited exactly to the needs of such progzostication, such as Mr. Wilson
keeps in Montana, will assist greatly to this end.

One important aspect of the project is suggested by the
question as to what would happen to these same farms if the present-
skilled overhead management and control were withdrawn; or if the
systems of farm organization marked out on these farms were adopted
widely over the state. No doubt a great deal of extension work would
be needed to make them succeed,- ahd there would be many failures.

Next in order to discuss is the extent to which the farm
reorganization Work which has been a part of the program of many
departments for many years, is entitled to be called experimental
work. Obviously it is more largely demonstrational than experi— .
mental. The‘departments do not dare suggest changes that they do
not feel confident will succeed. Such work is experimental, how-
ever, to the extent that it does verify or discredit, or suggest
modifications for further testing, in the ideas of those in charge.
This happens in enougi cases to make the experimental value of such
work worth while. That the demonstrational purpose of such activity
is parammt, however, is made evident by-the fact that it is commonly
looked upon as extension work rather than research. is generally
carried on, making more successful farms out of those reogganized and ‘
getting others to follow suit, figures more largely in the planning
of such reorganizations than does learning more about successful
farm organization and management. ‘

Recently, however, along with the current growth of in-
terest in experimentation in scenomics, a number of departments have
given thought to the research possibilities of such work, and pro-
jects have been planned in which the observation of the effect of
suggested changes and testing out ofidsas and principles, is
definitely specified as part of the researchprocedure. The in-
formation upon which the suggested changes are based is collected
by route methods in some cases, by survey methods in others, and
by loosely supervised record books in others — notably Illinois
and Kansas. Such objectives give direction to a larger portion
of the record-book projects than to that of W types. In
Kansas, for example, suggesting changes-M effects can be ob—
served seems .to be one of the major objectives of the project.
Other stations which have projects. under mp-énmdiinhjsuch testing

is an important factor are-Virginia, North Carolina, North Dakota,
South Dakota, Texas and Connecticut. Minnesota tried out a pro~
ject of this nature with-18 or 20 farms a number of years_ago. They
called them demonstration farms, however. Descriptions of most of
these projects will be found in Vol. II of last year's report on res
search in agricultural economics.

This new trend in farm organization research is of much
significance. It means new possibilities in the way of utilizing
data, and of cheeking the validity of various systems of analysis.
ReseaICh workers in this field can afford to give considerable
thought to developing the needed technique for it. Most of those
discussing it were conscious of the major difficulty that only
changes seeming assured of success are likely to be feasible, whereas
oftenthnes it is the more definitely experimental ones upon which
further light is needed. This difficulty is being met in a Way in
some cases by finding farmers actually practicing the propOSed system
or method, or believing in its pOSsibilities sufficiently to give it
a trial. Important problems need to be Worked out as to the kind of
records required for such trials, and for analyzing the resulting
data, and making adjustments for the special circumstaices of any yvar,
and eliminating the personal factor. ApproPriate methods of super-
vision need to be developed also. Several stressed the necessity of
allowing sufficient time for some of the important results to show
themselves - from 5 to 10 years with many sorts of changes. Another
difficulty mentioned was that the farmers frequently lacked the
capital to institute the changes suggested except very slowly. The
Fairway Farms project meets this obstacle by advancing the capital.

A number of them spoke of the unwillingness of many of the farmers

to accept suggestions from outsiders. This would indicate that the
proposal must have been presented to the farmers not as an experiment,
but as an improvement on his old methads — surely a Wrong approach.

In this connection, several felt that it would be necessary to own or
control the farms in Some way in order to get the desired changes
introduced and the necessary control of conditions — as at the Fairway
Farms. No doubt this would be true in many cases. The Fairway'
Farms project may easily denbnstrate that this can be done at no
financial loss, if the project is well administered.

None of the foregoing discussion exactly meets the issue
raised by Professor Weaver of Pennsylvania: ”my Own opinion is that
the Value of such studies would be doubtful since human conduct under
artificial Conditions is never normal,!and-a study of human actions
and reactions in such an artificial set-up, it seems to me, would be
of doubtful Value when applied to the general field."

Numbers 5 and 6 in the inquiry sent out, it will be re—
.membered, referred to technical matters of ec°nomio interest tried
out experimentally either by departments of economics alone, or more

322

particularly in cosperation with the natural science departments. The
projects reported under this head were in the field of production ~
feeding, fertilizer and soil control. The department of economics
had taken the initiative in getting such cooperative projects under
way. ‘ The various stations reported tht ~fiollowing technical pro-
jects of an experimental nature in the field of marketing: '

Effect of combine threshing on the storage of wheat.
Grading of samples of wheat by grain buyers.

Consumers' preference for apples in relation to quality.
Shipping lots of broilers and tomatoes to special markets.

Selling a surplus of local horticultural produce through
a special store.

Effect on sales receipts of grading peas.

Selling some produce through cooperative and private
outlets to see which netted the larger
receipts.

Selling lots of vegetables field-run and graded.

Selling in different kinds of packages.

No examples were mentioned) of experimental set-ups with
marketing business units; but two suggested that this would be the
only way of working out some of the problems of large cooperatives.
The Bureau of Agricultural Ecwomics looks upon one or trio of the

cooperatives which it helped organize as experimental ventures of
a sort.

A brief conclusion of the foregoing is considered from
the standpoint of research, there are some sorts of economic pro—
posals which can be tested out only by some sort of an experiment:
but for the most part, the method must be looked upon as supple-
mentary to the statistical method, oftentimes an extension or
amplification of it, the demonstrational values in such under-
takings are .often quite important, but experimental and demonstra-

tional objectives may complement each other if the project is
properly handled.

see

PART THREE

THE RESEARgH APPROACH.

The method of science, it has been sufficiently eXplained it"
an earlier section, is to isolate variations in single'causes and thEir.
associated effects. These Variations may be in time, fromwménth-‘to
month, year to year, or epoch to epoch, ordinarily called historical;
or they may be in space, from farm to farm, county to countv, or region
to region, ordinarily called geographical. There are other types of
variations which are not conspicuously associated with either time or: _
space in the physical sense, but which are highly sigiificnnt, such as ~
between business firms, or individuals, or institutions, or even nations.
To be sure, such entities have to be located in Space and time—somewhere;
but the important differences between them are not especially associated
with either. For example, it is not due to the fact'that one man lives
in south Dakota and another in MiSSOuri that one is bald and the'otfieé
not. A large part of the differences between farms are not associated
with Space factors in any way. For the want of a better term, this
third type of variation will be called non—geographical.

Once it has been decided by which approach an hypothesis shall
be tested, whether by isolating variations in time or space or "non- eo-
ra hicall ', the methods of collecting the data and analyzing them and

making inferences from them, are the methods that have been described

in Part II - statistical, analogy, case, informal statistical and ex- ' .
perimental. Hence it is not preper to speak of historical, gofegraphic
and non-rgoogrwhic as methods of research coordinate With the foiir'named;
One can, however, if one wishespnnke a new classification of research

methods on the ‘hasis of approach. That classification will be as
follows:

I Croes-section
A. Non—geographical
B. Geographical

II Time sequence-histo ri cal

a .11-. my

These could have been made the divisions of'our outline of“.
research method; but'if So, then we Should have had to discuss each of
the three final divisions above according to the five that were need.
It seemed to those responsible for the handbook that the significant
aspects of research method would be brought out much more clearly in
terms of the five used.

_. 324

rue
But it is alwayséne that no one basis of classification“
brings out all aspects of a subject. It is most decide‘ly true here.

from this second point of view; and out of this re—consideration t6
select for further discussion those-topics which the first point of”,
View did not sufficiently emphasize. The procedure followed in this
has been to as} four persons conspicuous for their approaches to prob-
lems from this ascend point of view to discussthe Subject of method
entirely independjtly of those writing from the first peih't of"view,
and then excerpt from their Contributions all no}: material bearing
directly on @thod. Dr. L. C. Gray'was aSked to discuss research
method from the historical angle, Dr. 6. E. Baker and Dr. 0.13. Holmes
from the geographical angle, and Dr. E. G. Nourse from the non-
geographic angle.

The task of selecting this material proved eirceeding v
difficult. Unfortunately for our present~purposes, gratifaf “
echelon; have differentiated themselves from their "renaxl'rei'ee his:
torians and geographers. Our interest in thahistorical and._.- .-
geographical apparach is solely as economists looking for evidence
as to economic hypotheses - not as historians or geographers en—

deavoring to build up a body of material in their fields.

A. Cross-section Analysis — Non-.Geogranhical.
Dr. Nourse's discussion of this subject provideshfor a" _-
statistical analysis of a cross—section whenever the necé'ssaryvdata
can be obtained, including even the "mathematical formulation of“ -_
certain mechanical connections which seem to exist"; and for iigraphic
presentation by which the facts are put into ingenioust devi seducom
tainers for display". But a great deal of crass—sectionanalysis
does not lend itself to such an attack. ' Dr.filTourse has desigated
the method needed instead by the term "qualitative description",
The discussion from this point on is in Dr. Nourse‘s own language
ilmost entirely:

v M - .._..__

Two general orders of qualitative description in'the invest—i:-
gational field may be differentiated; The first‘may be celied‘teetual
description and the second anal; tical description. Let us illustrate
both from the field of agricultural marketing. _ _.

We are concerned to know how wheat or cottonw-or‘ices are Slade.
The trite and ancient simply—and—riemand fomgla carries us kit's-tiny

distance on the road, since we must first know in what miner, findér‘
what. conditions, and subject to what varied influences defiends'do come
to price bids, and supplies are released as'ntarke't offers; ' Most of

the profesaors who discuss principles of price and the e“canomic i‘iincJ
tions of speculation could not answer ‘three out of ten questions as to

325

the actualities of future trading such as make up the routine detail
of thousands of Commission Muse clerks. The first step toward
analyzing the price—making process as it actually works out through“
such a market mechanism is to get a complete and accurate description
of the exchange machinery, together with. its ramifications to the
country shipping point and the ultimate consumer market.

Hare is an extensive array of facts which can be gatheréd
and authenticated and redmed to an understandable form onlv'through'
intricate descriptive account of what is actually taking place in the
grain or cotton market. It involves not judgment as to how” these
transactions affect price hor of the economic desirability o'f'the
prices attained, or the methods by which they are arrived he.” Tn .
the main, economists have been too lazy to defer the fonnulationmof
large theories of price until after such minute. descriptive investi—
gationshad been carried thdpigh Wei—iiivng-rtoiaketheirfienersl—
uationsihomgh or unwilling to take their generalizations through
the laborious process of deductive applications to such intricate .—
concrete stages to see how fully their principles explained the;
facts or were consistent with them. The research worker who can.
supply this gap in a thorough and trustworthy manner requires a
higc-order of mfisnzmship, and his handiwork should command the
resPect of all his fellow craftsman.

This, however, is not the final and of research. we
desire not merely to know what is happening in the market in the
sense of learning pmcisely what transactions are'carried on and—
upon what terms. '39 wish also to examine the produéevof the market:— _
mechanism in terms of prices. Hence the next step, once our factual
data are in hand, is to proceed to the task of analysis in order that
we may see how these different transactions result in establishing
prices in different instances at one level or another, and how these
countless and often quickly struck prices fit together into a. price
system, eXerting certain pulls and pressures upon each other, in—
ducing certain aberrations from or returns toward more general trends
which prevail for longer or shorter periods.

The first attempt at analyzing the operation of the market
will consist in obserVation of a. great mass of actual prices with
the idea of giving a descriptive account of the price system in terms
of grade differentials, place differentials, and price differentials
that We may have a picture not of market trading but of market prices.
Once we have this picture of prices descriptively developed out of the
inherently meaningless dollars and cents figures, we then prepared for
the next step in the process of analysis, namely, a statement of the
issues which arise from contemplation of the price situation descrip-
tively set forth. Do grading methods affect prices in this or that
way? What is the influence of particular transportation and storage
conditions? How do future trading Operations affect cash prices?
Has the Grain Futures Administration impoverished the farmer? These

a picture reduced to manageable size and given clear outlines througi
and application of two orders of qualitative technique.

Here, however, and even beyond this point, in the actual
formulation of theories the analysis, however much it may employ the
calculating machine and the devices of statistics, retains a strongly
descriptive character, since after all what we seek is a verbal ex-
planation of what is transpiring, and more figures can enlighten and
convince us only as they secure an adequate setting in a descriptive
account ofthe economic phenomena from which these numerical ex-
pressions are derived by a process of measurement.

The progress of science has been due to the development of
a great number of trays of examing natural and (more recently) social
phenomena and of analyzing and interpreting the results of this ob-
servation. If we turn to consider the variety of methods used by
workers in the older investigational fields, that is, the so-called
"natural sciences," we will find that qualitative descriptions has/e
held and still do hold a position of good repute and great service—
ability. It is by accurate observation of a variety of natural
phenomena, classification of these facts, inductive reasoning from

them, that the older sciences have blazed the trail toward the
truth. ‘

It Was only a relatively late stage in the investigational
process in biology, physiology, pathology, and therapy that sta-
tistical method has come in as an aid to analyzing a. satisfactory

result of previous studies of a qualitative character. In a.
manner essentially analogous, economic investigation into the struc-
ture and functioning of economic organisms has given us such under-
standing, as we have of the actual processes by which economic life
is carrying forward today. Not alone because of the inadequacy or
inaccuracy of these previous studies, but likewise because of the
continuous process of eVOlution taking place in our institutions
and economic activities, it is important that the process of pain-
staking obserVation and analytical description should continue and
our methods in this field be constantly examined and, so far as
possible, improved.

Like all other methods of investigation, this is subject
to abuse and is frequently regarded with some contempt by those who
ply their trade with other tools. It is thought of as a method re-
sorted to by the lazy or superficial. In spite of the laziness and
superficiality of many vino have written photographic and uncritical
descriptive accounts of economic phenomena, the real point is not
that such method is inferior or superior to others, but rather that

” 327

it occupies a different place in the research economist's tool kit.
Generally speaking, it should be regarded as the first tool with which
he will attack his problem. ‘

That is to say, if we have a problem that is one of activities
and. relationships taking place at the moment or within a very short
period of time, the first thing is to ftrrmilate this problem, identify
the isaues, see what are the detailed questions to which answers are
desired. It is in this field that qualitative analysis functions,
undertaking the task of ascertaining just what is happening and how
the several actors or mechanisms are severally performing in the
producing of this total situation,

Such a formulation of the problem may prompt a flirther in-
quiry as to the evolutionary proceSs by rflnich this situation came about,
and this may be indiSpensable to any thorough Imderstanding and wise
action with reference to it. If so, the method will shortly shift from
that of analytical cross—section to one of historical examination. On
the other hand, and quite possibly at the same time, the statement of
the nature of relationships within an economic situation will proba‘blv
reveal a greater or less number of detailed issues which n’dmit of quan—

titative measurement, and for which the statistical technique offers
invaluable assistance.

Too much emphasis, however, cannot be laid upon the fact
that the egregious shortcomings of many statistical studies in the
past have been rather evidently due to the inadequate formulation of
the problem to which the statistical technique has been annlied. This
has resulted in the selection, acceptance, or use of numerical raw
materials which were utterly inadequate or inappropriate to the deter-
mination undertaken or to the acceptance of inrpliCations from quantita—
tive results secured, whose untrustworthiness or even absurdity would
have been apparent to anyone who observed closely the functioning of
the elements whose quantitatively stated attributes or nroduces had
constituted the raw materials which wont into the hopper of the sta—
tistical machine.

Every research worker therefore should aspire to include in
his training the development of as great proficiency as possible in'
handling the qualitative description phases of research; Likewise,
in the planning and execution of every research project, searching
scrutiny should be bent on the work to ascertain that the descriptive
method has been used in those places where it was called for and

supplemented by other techniques where they could bring; added re-
sources to bear, and that wherever and whenever qualitative de-
scription has been used it has been carried out in a workmenlike
manner and the accuracy and credibility of the product has been care-
fully checked with reference to the whole project.

‘328

It is important at this'point to distingtfi'sh'hetween de— '
scription as a method of analysis, which is research, and description
as presentation of the results, which is rhetoric. The objective
of qualitative description as a method of analysis is to find out what
elements or parts comprise )what is being studied, and how they work'or

function together. The objective of description as rhetoric is to
tell about the foregoing.

Dr. Nourse's discussion makes it clear that the handbook
nowhere, not even in this section, is as specific as should be as to
Just how qualitative description is performed. There has been talk
enough of the need for qualitative analysis, but little said as to
her: it is done. (See the section on the "Case Method" for the most
specific detail on the subject). How does one determine that
elements constitute a structure or organization or institution? and
how does he prove his conclusions? No doubt some discussion of
the testing of the authenticity of Various forms of evidence would
be pertinent here. No doubt the principles and prficedure of logic
are closely involved. It is obvious from Dr. Nourse‘s discussion
that frequently the evidence of a number of similar sources, or
from a number of cases, is to be combined. Hence our discussion

of informal statistics becomes pertinent. The answers to many
of the questions posed in analytical description call for use of
data that are at least informally quantitative.

Examples of data of a largely non-geographic character
which are commonly handled by statistical, informal statistical,
and case methods, are local data as to size of farms, farm build»-
ings and equipment, and working capital; age, education and ability
of farmers; size of farm families. As between larger divisions
of a country, some of‘these, it is true, may take on an important
geographic character; but not local variations in them. So is it
with most of the variations between marketing units. Geographic
factors are those which are definitely associated, either directly
or indirectly, with other factors which have a definite IOCation -
such as climate, soil, topography, rivers, etc. Geographic in.-
flusnce is a matter of degree. It determines very largely where
coal will be mined, to a lesser degree where corn will be grown;

still less where it will be eaten as the staple human food: and to
only a limited extent There people play a certain kind of ball. To
be sure one can find indirect traces of geographic influences in al—
most anything, perhaps even in the Boston custom of eating beans on
Saturday night. But this does not Warrant identifying geography
with the study of civilization as some geographers are prone to do.
Everything has to have location somewhere. But geographic factors -
climate, soil, topography, etc. - my have had little indeed to do
with determining it.

_, 329

B. Cross—section Analysis - Geographical.

Dr. Baker's contribution to this subject consisted We.
careful claSSification and description of the purpose, scope and
method of all the studies made in the United States with any signifi-
cant geographic content. What follows is an attempt to select from
this in the order given all the statements relating to method. Some—
times these are given in Dr. Baker‘s own language, sometimes translated
into the language of the handbook. The nib-committee is responsible
for this translation, not Dr. Baker.

. The first type of methodology to be mentioned is of course
the simple listing or mapping of conditions, physical or human or both,
by locality or region. The method is often that, of the atlas or the
gazetteer, and may include both. With reference to agriculture, it
takes the form of a compilation of facts, commonly statistical, and
presented on maps to assure completeness and accuracy, as well as in
tables and text, relating to temperature, moElture, topographic and
soil conditions, the use of the land for crops, pasture, forest, and
non—agricultural purposes, accompanied by notes on the relative import-
ance of the crops and classes of livestock and the systems of farming,
closing sometimes with a. description of the racial, religious, and
other characteristics of the people. Frequently a little interpreta-
tion is intermingled with the mapping and listing,~ usually a note as
to the climatic or other physical Conditions that have favored the pro-
duction of a crop or the development of a system of fanning. This
represents the introduction of geographical analysis. In the studies
on a national and world scale, as in "The Geography of the world's
Agriculture“, the production data have been mapped by Counties or
similar geographic units.

Along with the mapping have developed various geographic
devices for use in these maps, such as dots, circles of Various sizes,
cross—hatching, perpendicular bars, etc., mich are discussed in a
special section of the handbook.

It is not always appreciated that these geographic devices
are based on coefficients, and that the choice of them involves the
principles relating to coefficients given in Volume I. Thus a dot
for 1000 acres of com drawn to the same area scale as the sap itself
assumes that corn acreage should be related in a coefficient to total
area, not to land in farms, or land in crops.

It should be obvious that mapping itself is not geography -'
it is merely a way of recording or presenting data so as to make it
readily available for any sort of analysis. Geographers, because
localization is what they seek to explain, have more use for data
presented in this form than most other scientists.

The tiichigan Land Economic Survey is the best example of a‘
descriptive local field survey. The objective of this is an inventory
presented mostly in the form of maps, of the present nature and extent
of the land resources in soils, forests, waters, etc. This includes
present land use. Following is a description of the method of field
work:

"We aim in our mapping to be so exact as to location that the
number of acres of each different soil type or cover type to be found
on any particular forty or twenty-acre tract can be read directly from

our maps. With clean napping sheets in their notebooks, a crew of
two men is assigied to the mapping of an area of about nine Square miles.
One man maps base data (culture and water features), soil and lay of the
land, while the other man maps base data and lend utilization, forest
and non—forest vegetation. Before starting work, the men are informed
as to the location of control points which have been previously located
or reestablished on their area. They being their work at one of these
control points (which is usually a section or quarter section corner)
and gridiron their area by running lines north and south or east and
west along the section and quarter section lines, mapping a quarter of
a. znile on each side or their line or travel. By this method, the same
degree of accuracy is maintained ‘in the difficulty accessible cut-over
or timbered areas as when mappi ng along a main travelled highway. Nine
square miles of mapping has been found to constitute an average of three
days'_ work for the two—man crew."

"Intent in land ownership, assessed valuation, taxation, public
expenditures, history of occupation, agricultural and industrial history,
and the county's business and production, are similarly reduced to maps.
[From the records of the local abstract office, a map is made showing the
present omership of all acreage property. The 'intent in land owner-
ship' of each parcel of acreage property is obtained by interview with
the local township assesaor, as he is the person usually rest familiar
with the plans and policies of the owners in regard to their land.

After this interview the intent in ornership is ‘classified and plotted
on a. map. The assessed valuation of each parcel of land is copied
from the latest tax rolls and is recorded at once on a map. Later
this valuation is classified as 'under $3.00 per acre‘, ‘$5.00 to $9.00
per acre', '$9.00 to $21.00 per acre‘, etc., and plotted on a map.
Local trade areas tributary to each shipping point are outlined on a.
map, the boundaries of each area being based on such factors as topo-
graphy, road conditions, distances, mail routes, land ownership, and
local favoritism. The corn y's business and production is related
to each of the trade areas chem on this map. The data on local
tax rate, public expenditures, and all historical data are also tied
in to a. definite location on the maps. Because our data, both
physical and economic, are located on maps, a direct correlation be:,.r " '
4 ..~ Maxim 'nmre-.fac.t;nrr.._ {11"}: as --soiLanr‘...m an: sw‘ grail IL 4' 3/25“”

331

or between intent in land ownership and tax delinquency, can be readily
made."

Clearly the only unfamiliar technique involved in this is
that which relates to the method. of working in the field. In their
present form, the results of this survey are available for geographic
analysis, land utilization analysis, or any other kind of analysis de-
sired. They are the raw materials of geography, not geography itself.
The technique of collecting and mapping such data is not the special
property of any field of knowledge.

We now come to the analysis of geographic variables, that
is, the relating of different conditions to each other by locations.
About the only hit of methodology used thus far in most of such an-
alyses is to put the two variables on the same or similar maps, look
at them and see to Tfl’lat extent and where they coincide or fail to
csincide. From this, one my sometimes draw lines showing the ap—
parent rainfall, temperature, growing season and other limits for
particular crops. Mr. Bradford B. Smith's article in Vol. 30 of
the Journal of Agricultural Research (p. 1083) illustrates an attempt
to apply formal correlation technique, discussed at length earlier,
to this problem. Apparently two different lines of attack are being
followed in such analyses. The one followed by geographers is to
observe what crops are being grOWn in the region and relate this to
the best long-time averages available as to rainfall, temperature,
etc. This assumes that the practice of the farmers may be taken
as sufficient evidence as to the tolerance. The other attack,
illustrated by E. O. Wooton's "Settlers’ Progress in Dry—land Earning
in Eastern New Mexico", as well as by Smith's study already mentioned,
and several others, is to get behind actual practices of farmers to
a record of their experiences in the matter of rainfall, frosts,
yields, crop failures for various reasons, and economic progress
generally, and analyze the relationships more closely, often by formal
statistical technique. The assumption behind this sort of analysis
is that present practices may not be properl; pted to conditions.
This of course takes the whole problem outside the realm of geography
into farm management and land utilization. Analysis of this sort
is being based largely on farm-to-farm surveys, of the same general
sort as used in farm management and detailed land utilization studies,
and is being made by specialists in these fields.

The analysis of shifts in production and the like and reasons
for them from deCade to decade, of course represents the historical
approach and not the geographical. Knowledge of such shifts, how-
ever, is fundamental to geographiCal analysis. It is necessary to
lmow whether margins for crops have remained more or less stationary
from decade to decade before one can isolate the influences cf 8. more
permanent nature such as due to soil, climate, etc. The most favorable
climate and soil is probably to be found in the areas where dense pro—-
duction has continued for many decades. Here is a case of the using

r

,532

of the same data in two sorts of analysis, crossmsection and time-
sequence. So far as the cross~section analysis is concerned, the
only methodology involved seems to be making maps at different per-
iods and observing to what extent they resemble and differ and where.

Professor Wellington D. Jones of the University of Chicagp
has developed a code for recording observations in the field in local
studies. Each use of the land, as to whether cropped, pastured, idle
or farmstead, and each particular crop use or grade of pasture or other
use, is assigned a digit and position in the numerator of a fraction,
and each natural site characteristic — slope, soil, profile and drain-
.age - a similar digit and position in the denominator. i.e result is
a fraction, such as §§§ which can be placed on a map in the proper

4~ ,
10Cation. From this, any desired number of types of maps can be made
in the office; A method used by Professor Jones for correlating use
and site characteristics is to construct a map in which a different
color is assigned to various combinations. Thus, let us say, all areas
are colored red in which corn Wes grown on Slope l or 2, unth.Soil Pro—
file 1; all orange in Which corn was grown in Slope l'or 2 with Soil
Profile 2; etc. In such a study in Illinois, eleven different colors
were made to represent all the Conditions of use and site characteristics,
These color areas Were then measured by a planimeter and reduced to per-
centages. Similar methods were used in mapping and analyzing and site
,and use in the village in the center of the area.'

The following statement by Professor Jones indicates the

. reason for the foregoing technique: "The clearer the complicated pat—
terns of the land use and of factors bearing on land use are set forth,
the greater will be the likelihood of geographers finding out why areas
are used as they are, their suitability to present use, and their possible
'and probable future unset Not the least of the ways by Which causal
relations between these various phenomena are discovered is by noting
repeated coincidence in distribution between different phenomena, and
maps supply the best tool for recognizing repeated coincidences in dis—
tribution."

Clark university has made a field survey of a section of
western Massachusetts including the Connecticut river valley in which
the effort has.centcred around the making of celered maps showing land
use, surface features such as flood planes, deltas, terraCes, moraines.
Certain type farms were selected and mapped in more detail.

_ _ The regional studies of the sort being described by Professor
Carl 0. Sauer — illustrated by the Kentucky GeolOgical Survey studies
seem to differ in no important respects as to methodology, but have been
,broadened out in scope to include the whole of the civilization and cul—
ture of the region. ' ‘

333

The studies being made_in Connecticut are directed at showing
in a thoroughgoing way the effect of soil type on the agricultural
economy and living conditions in the area. Most of the data are being
collected much in the manner of farm managewent surveys. Soil types
are mapped out carefully by the soils division. In the more recent
studies, areas have been chosen which a e typiCal of'the Soil types
whose effects are to be observed and fangs chosen at randmi in these
areas, and 'n addition a Careful soil map is being made of each fora.

Other problems in method illustrated by these studies are
those 0f the clausificition of land according to present and potential
uses, and the districting of a territory into SUE—regions, types of
farming, etc. The information provided, however, does not indicate
any procedures not outlined elsewhere in the handbook. The difficult
feature of all such analyses is the choice of a basis for division and
locating of the boundary lines between classes and zones.

Professor Holmes after reviewing the course of development
of geographical analysis in the United States, Concludes as follows:
"The foregoing summary of the work in this country and abroad seems
to indicate that the geographical approach to agriculture up to this

.time has, for the most part, been descriptive rather than analytic.
The procedure has run from the presentation of the actual facts with
reference to agricultural production to the graphic presentation of
these facts; and thence to an effort to differentiate the agriéultural
development of the given country on the basis of differences to be
observed in he actual production program. There are, to be sure,
the beginnings of a more refined analysis, and it is to be hoped that
this more intensive analysis will becoin the dominant feature in the
Work immediately ahead."

Professor Holmes suggests that the geographical approach
should be considered as upon three planes, the lowest, the descriptive
and graphic, with which this present account has already made us very
familiar; next, the sinthetig, in which the effort is to combine in~
dividual oSpects such as craps, livestock, physiCal features, into
'cemplexes such as zones or subwregions, systems of farming, economics,
etc., and third, and highest, What may be termed the analytic. As to
the first two of these planes, ProfGSSor Holmes writes: ‘"A great deal
of.practical utility characterizes the results on these two planes of
research. The results yield information as to competing regions and
countries, as to sources of supply of various commodities, and furnish
a background of information and understanding with reference to-the
agricultural industries. Tremendous irportance was attached to the
results of this sort of research_which Were available at the time of
the Ear. Perhaps no other single publiCation Was so frequently re»
ferred to in the days of our efforts at producing food for the support-
of our military troops and those of the allies as Was the "Geography
of the :orld's Agriculture" fihich was putlished at about that time by
the United States Department of Agriculture. The policy of the na—
tional and state food administrations nos probably almost entirely

834

built upon information supplied ty this Sort of research. It is quite
clear that the results of such studies are absolutely necessary as,a
starting point for the tore thoroughgoing analysis which should by all
means follow.”

Here follows ProfeSsor Holmes discussion of the third plane
of approach: "This plane involves going h 70nd the mere presenting of
individual elements in any agriculture. It goes telow those things,
into the causes which account for the agricultural development as it
is found. It necessarily takes the in ostigator into an analysis of
the natural agriculture reSources in their physical and technical re—
plationships to plants and animals. It must take him also into the
nature of economic conditions and forces as they impinge upon the
agricultural industry of a given region, and together with the natural
conditions and forces, give it shape and character.

"At least a heginning of the analytic method is to be found
in almost any study such as has been described atove. The physical
demands of crops, for eXQiple, are repeatedly cited as explaining the
boundaries of our great agricultural regions such as the Cotton Belt,
the theat Belt, and the Corn belt. Rainfall, host and cold, length
of growing Season, character of the surfice and the soil have all
been brought into Consideration in an explanation of uhy the farmers
of a given region devote their resources to the production of a given
crop or group of crops. Similar orplanations of the distribution
of livestock products and the technical nature of the industry as it
differs from region to region have been advanced. Likewise, the
general influence of market location and transportation facilities
have been cited as eXplanation for the type.of agriculture which has
been found. Closely akin to this is the attempt to account for trade
route and market centers in terms of the qualitative analysis of the
distribution of population and the occurence of non—agricultural re—
sources which have been the cause in part, at least, of the development
of industrial center which are at the same time consuming centers of
agricultural commodities, -

‘"Here, as in other fields, we are feeling the need of a more.
refined and efficient type of analysis in terms of quantitative measures.
The use of'refined statistical methods in analyzing the forces back of
the development of regional differentiation in the utilization of agri-
cultural reSOurces has been slow in developing. It appeared much earlier
in fields Where its se could be the more easily developed, such as in
the factors determining the profits to be derived in farming. However,
the some considerations which make necessary quantitatiVe rather than
mere qualitative analysis in these other fields, are evident in the
fieLi of economic geography, and beginnings are being made in the use
of Various correlation procedures in order to determine the relative
effect of the various forces at pork; Nothing is more evident than
the fact that the agricultural development which has been reached in
any given region is due to a complex set of forces and obnditions and
that very great errors of judgment may be made.as totthe relative im-
portance of any given force or group of forces unless resart is had_to

I7

:3 335

quantitative analysis. The writer wishes here to refer to an ac~
companying report by Mr. E 3. Hurd cf the methods of quantitative an—
alysis being used here at the Iowa Experiment Station in our type—cf—
farming study. All of the shortcomings and pitfalls which have already
been discovered by the users of correlation method in other fields of
research must be guarded against in this field alok. A blind adherence
to the results of numerical analysis without the neceSSary interpretation
would be quite as misleading and disastrous in this field as in the field
of price studies.

”Our work here at the Iowa Station has revealed to us one con-
sideration which perhaps has tended to be overlooked in most geographical
analysis as applied to agriculture. This is he outstanding importance
of entrepreneurship in the determining of that are agriculture actually
is. The analysis must by all means proceed from the motive of the
individual farmer in his efforts to obtain maximum utilization of his
ovn resources. It is he who is actually making the adjustment to the
natural environment of which his limited holding is a portion and to the
forces and conditions which make up his economic environment. We feel
that_there is no more important point for emphasis in this whole prob em
of research method in this field.“

Here follows a resume of Mr. Hurd's description of his method
of studying types of farming geographically:

The first step Was to obtainclata as to individual farms in
the territory being analyzed. In Iowa, these are available in assessor's
reports. It was obviously inpOSSihle to include all the farms in the
territory; hence a random sample was taken. The next step was to work
out a scheme of classification of farms and to sort the farms in the
sample according to these classes. This classification can be on the'
basis of individual crops or livestock products, or all combined into
various complete systems of farming. We were then ready to undertake
our central problem, namely to observe the forces and conditions that
put each of these farms-into its class. These forces or conditions,
we have classified as (l) geographic,~or land in the economic sense;
(2) natural conditions other than geographic - plant and animal pests
and diseases; (3) technical development; (4) purely economics, such as
prices and costs; (5) personal.

336

Since Wefiere interested in discovering primarily what con—
tribution.the geographic forces have made to the different types of
farning, it tecomes necessary to hold the influence of the other ferces
constant while we observe how the variations in geography affect the
fanus_under consideration. For any small area, such as Iowa, one set
of prices and Costs tends to preVail during one year or interval of
tine for the entire state. The s;ne is true of any form of technical
development. Any new discovery is rapidly made ayailahle to all the
people through the press and comnercial agencies. The non—geographic
natural influences may be divided.into two groups, those which are well
eStahlished in the whole area under study, and those which are estab-
lished only in certain parts of the area or bid fair to Iecoming im—
portant some time in the future. If the former are workiné “1th equal
intensity all over the area studied, the Variations in farming within
the area probably may "e assumed_to be due to other causes., A PeSt
which infests only part of the area-under consideration may he treated
like the geog‘ phical faCtors, except that one will need to estimate
how soon it will.reach the rest of the area and what will be the prob-
able effects in light of the nature of the pest, the geographical fac-

~

tors in the non-infested regions, and the aVailable methods of control.

to are now ready to separate the personal conditions from the
geOgraphic. He She 1 illustrate this nith the case of corn. How
much of the variation in corn acreage is due to the personal elements
and how much to the geographical factors? In order to throw light on
this question, we selected one township where the natural conditions
were as nearly uniform as it is possible to rind them. This township
was selected from a group of townships that showed township averages
having very little disPersion. This in itself suggests that the
Variations in natural conditions within the tormship were not large
enough to have economic significance. In the second place, his town-
ship was selected where the known geographical variations were at a
minnnmn, as evidenced Ey soil reports.and first-hand information. We
can assume, then, that for all practical purposes the differences in
the geographical factors will not affect or influence differences in
type of farming within this township. The corn acreage in this town—
ship in 1984 averaged 41 per cent. The standard deviation from this
average Was 9.5 per cent. This means that about one—sixth of the
farms in this toymship had less than 81.5 per cent of their land in
corn, two-thirds of the farms, between 31.5'per cent and to 50.5 per.
cent, and the other one—sixth of the farms, in excess of 50.5 per
cent. This Variation is due primarily, if not entirely, to varia-
tions in the personal attrihutes and conditions of the individual
farmers. * ‘ ' ‘ '
To get a measure of the geographical influence, we averaged
the percentage of farm land in cern'for.one year in each region of the
state that has uniform natural conditions. It seemed probable that

 

*The question may still be raised whether the land may not have
varied considerably from farm in each tonnship, although sinilarly
in each township..

339

TABLE II

The Relation Between Soil and Crop Production in
Blackhawk County, Iowa, 1925

 

 

: :Yield :Yield : ProduCed per : Ratio of
Soil : Topographic of : of : 100 Acres of : Oats to
Area : Position :Corn :Oats : Farm Land : Corn
: :ner A :ner A : Corn : Oats : Produced
-Iowan Drift : Upland : 47 : 42 : 1504 : 882 : 59
Iowan Drift : Upland & : 46 z 39 z 1380 : 663 : 48
' : Bottom : . : . : z :
Mississippi : Upland & : 45 : 41 : 1395 : 779 : 56
Loess : Bottom : : : : :
Mississippi : Upland : 53 -: 47 : 1908 : 987 : 52
Loess : : : : : §

 

Table III shows the average percentage of land in corn in Iowa
for each of the five "soil areas". The standard deviations in the third
column of this table are of some significance. The Wisconsin drift soil
area has 40 percent of its land in corn, and furthermore the townships in
this area show less variation from this average than the townships in
other areas. All the standard deviations are rather high, even for a
state with as nearly uniform soil as Iowa. They would be still higher
with individual farm data. '

TABLE III

Effect of Soil Areas on Per Cent of Land in Corn-
Township Averages

 

: Per cent of Per cent of the:Standard Deviation

 

 

Area_ - :the entire state: area in corn : of Townshins

State : 100 : 33 : 8.8 ‘
Mississippi : : :

Loess : 15 : 29 : 7.9
Southern Iowa : : :

Loess : 12 : 26 : 6.8
Missouri Loesa : 20 : 42 : 6.5
Iowan Drift : 12 : 28 : 5.8
Wisconsin Drift : 22 : 4O : 5.5

o.

 

340

. The influence of surface upon types of farming cannot be told
in the Sane .Iay as soil. This is because direct infonnation on surface
conditions has never oeen collected, A topographical map of Iowa has
never been made. T0pography, ho'.ever, may be neasured indirectly by the
percentage of tillable land. e have never taoulated the percentage of
tillable land on a to,nship oasis. The range in this factor between
counties varies from 53 to 95 per cent. The correlation between till-
aole land and percentage of land in corn is .78.

.. ‘ IOWa is a poor place to traCe the influences of climatic con-
ditions upon type of farming. The variations in climatic conditions

are small and hat Variations do occur are usuallr lacconmanied with vari-
avions in other geographical conditions. Climate is reflected, however,
in t1o other variables we have us ed, namely, the ratio of corn yields to
oat yields, and the percentage of corn merchantable. The first Was de—
rived 0y dividing the ten— —year average yield of corn in each county by
the same ten— —year average yield of cats. The percentage of merchantable
corn is rive-year_average of the government crop reporters' estimate.
The ratio of Corn to oat yields has a much higher correlation with the
percentage of land in corn than corn yields alone. This reflects the
fact that comparative advantage is more significant than absolute ad-
Vantage. The coefficient of multiple correlation of county averages,
using as the dependent variable the percentage of corn, and as indepen-
dent Variables the percentage of land tillable, the percentage of corn
merchantable and the ratio of corn to oat yields, was .84. Climatic
factors include many different variables. In states where variations

in climatic conditions are more us rked than in 101a, more careful methOde
will be needed than the foregoing. Such climatic factors as rainfall,
length of growing seaSon, temperature, and winds in the area should be
analyzed. Perhaps it is no accident that the line marking where oats
yields and corn yields per acre are equal in love follows closely the
l50-day growing season line.

In conclusion, the foregoing discuSSion of geographical an-
alysis can perhaps be said to amplify our discussion of research method
in the follordng particulars:

1. Special details of procedure.in collecting certain
sorts of data in the field and in field mapping.

2. Special emphasis on mapping as a method of recording
and presenting data for inspection; and some
special geograpr ic devices.

3. A crude form of association analysis based on observa~
tion of coincidences or lack of coincidences in
maps for_different factors or at different periods.
The logical basis of conclusions based on such
methods are not examined to any extent here.“

337

the variations.in corn acreage that are due to variations in the per—
sonal factors;will average out in such a region, and that the remaining
Variations will be due to differences in geography Letueen these‘oroas.
The farmers in each area till be trying to get the maximum profit for
themselves. The individual farmer will err consideraoly o;nng to
differences in his efn personal aptitudes and Conditions; but most of
this error will prohahly not he revealed in the aveIECCS Of the areas
since the highs and loxns will-offset each other. The average in any
one area will probably not coincide exactly with the flOSt profitable
amount Quin; to economic lag Eetween he time'uhen price changes occur
aid changes in method become hnonn era the farmer's actual adJuStflent

to them. Neither will there Be a prompt adjustment to the test—knotm
ways of dealing with some new plant or animal pest or diSease. However.
it is possitle that the various areas will have about the Sate amount
of economic lag between the average type of’farming and the most PTOfit'
a:le type for that area.

The areas selected for averaging must have uniform geographic
conditions, and at the nae time to large enough so that differences in
the personal factors anfl conditions may average out. Probably a town—
ship is large enough area for this. The standard—sized twmship in
Iota contains about 144 farms. But geographical conditions vithin
most totnship are not as uniform as the tonnship used atove as an il-
lustration. To the extent that this is true, Variations in farming
due to geography will be confused with variations due to differences
in the personal factors. Ln area smaller than a tounship could not
very roll be used owing to last or statistics, and even if it could
there would be danger of having too small a sanple to be representative-
In case larger political units than townships are used as a basis for
analysis, more geographical factors will be obscured than in the case
of using single tomnships. The standard deviation of percentage of
land in corn for 1925 in ions based on township deviations from the
state average is 8.8 per cent on county deviations from this same
average, 7.7 per cent.

The geographic factor will also help determine the size.of ,
the factor taken. If it is climate, a county orjafstate will bela
small enough unit to obtain a fair degree of uniformity within the
unit. The larger area for tnis purpose would have an advantage over
the smaller in that Variations due to geographic factOrs other than
climatic conditions would tend to be averaged out as well as the Vari-
ations due to the personal'factors. On the other hand, if the analysis
is concerned primarily.with surface and soil factors, the smalleSt
political unit is almost a prerequisite to successful analysis; and
surface and soil factors are the chief geographical reasons for varia-
tions of types of farming Within small areas.

The soils of the United States have been mapped into thirteen
maJor soil provinces or regions. ~Liany experiment stations have recog-
nized that these soil provinces or regions contain elements too diverse
for their purposes, and consequently they have divided their states into
”soil areas." In Iowa there are five such soil areas. The soil area

f 338

contains many ”soil types”, each one of which may be found in Widely
.Separated parts of the area. It is seldom that a single square mile
can be found that will contain only one soil type, although a soil

type may amount to over 50 per cent of the total county area. A

soil type is the smallest unit of classification. It is a soil

which, throughout the area of its occurence, has the same texture,
COlor, structure, character of subsoil, general topography, and de—
rivation. If one is studying wide variations existing betWeen differ—
ent parts of the United States, one mar observe the correlation between
soil provinces or.regions and types of farming by using state data; if
one is studying variations existing in different parts of a state, the
correlation will be made by using county or township data. If the cor—
relation is Eetween types of farming and soil type, one should use
individual_farm data. Only in rare Cases may a whole township be found
which will have a sufficiently high percentage of one soil type to
warrant the drawing of conclusions on that assumption.

Blackhawk County, Iowa, however, can be handled on a township-
average basis. The Cedar River, which divides the county into equal
haIVes diagonally, also Splits it into two distinct soil types, each
divided into,tto phases according to its tOpographic position, a river
and an upland phase. Tables I and II give the comparative figures in—
dicating types of farming. In most cases, individual farm data would
be necessary to show the same differences.

The Relation Betteen Soil-and Land Utilization
in Blackhamk County Iowa, 1925

 

: : ' Percentage of Land

Soil ' Topographic : in Different Uses.
‘ Area Position : : . V .
: Corn : Oats :Wild Hay:Tame Hav: Pasture

- , n
o o ,

.

Iowan Drift Upland : 32 ; 21 : 3 :' 10' 27

Iotan Drift Upland & : 30 : 17 : 2 : 8 z 33
‘ : .Bottom : : z : . :

dississippi : Upland & z 31 : l9 : 8 : 9 : 30
Loess : Bottom : : : z :

Zfississippi : ‘Uplend :A 36 : 21 : l : 9 z 35
Loess ; : : z ; :

 

)s’)’

.1fl,lgv415',vnl'hr,34l

4. lapping and measuring from such maps the areas in Which
' ‘ such coincidences, or the-lack of them, occur —
Professor Jonels procedure.

5. The SYnthetic process described as a phase of geographic
analysis, including such things as setting—up systems
of farming, zones and the like. It is not to be
understood that this is peculiar to geographical an-
alysis. It is merely that the handbook has scarcely
mentioned it heretofore.

6. Some Special details as to methods of isolating the effect
of the geographic types of Variables — as in the
Connecticut and the Iowa studies.

So far as definite correlation procedure with such data is
concerned, three earlier sections of the handbook, ”Association Analysis",
"Elementary Correlation AnalysiS", and "Correlation as Description of
Geographic Data" reach further than anything developed in this section.

. As both.Dr. Baker and Professor Holmes have stated, analytical
”methodology with geographic data is still not greatly developed. It is
'important that it should be developed further. The geographic surround—
ings of current problems are as no essary to good perspective as the
historical background. Geographic conditions alone cannot determine the
future developments of an area; but they do impose some hard limitations
on them Which need to be clearly recognized. Programs of Southern div;
ersification, of local adjustment of productiOn and consunption, of con-
trol of production and adjustment to foreign competition, of lOCal pro-
duction readjustment, as Professor Holmes points out elsevhere in his
paper, need just such_geographic analisis as a prophylactic against
wild proposings. The geographer himself is not in a position to fore—
cast future developments or plan reorganiZations; but his fasts and
his relationships are among the most important bases of such forecasts
and plans.

C. Time Sequence Analysis - Historical.

Ideal historical analysis would consist of a comparison of
'a continuing series of cross—sections with a View to noting the changes
between these and discovering the relationships between the various
changes. Given data in this degree of Completeness, it would surely
employ formal statistiCal method, as in fact it does in dealing with
many aSpects of recent periodS.’ Lacking such data, it resorts to
informal statistical methodi lacking this, to case or monographic
studies and even analogies. Once we pass out of the recent period
when statistical records are aVailable, the whole problem of historical
analysis takes on Special characters because of the nature of the avail-
able information. The method of working with such material represents
a contribution to methodology that has gone further than anything pre-

342

sented elsewhere in the handbook. Herevith follows Dr.L.C. Gray's dis-
crssion of this pspect of methodology together with a discussion of the
process of sinthesie, described fer geographic material by Dr. Holmes,

as it works out tith historical material. The extreme breadth and-
variety of the data,xmich history as a subject conprehends make synthesis
a highly significant part of its methodology:

It has been customar" to recognize certain important sub—
divisions of historical method. These include the finding of the
sources or process of bibliography (Quellenkude); the critical testing
of the origin and genuineness of the scarce and the validity of the
facts contained in it, together with the arrangement of the facts thus
established (Kritik, operations analytique); the interpretation and
grouping of facts in their various relationships. (guffassung); and
the formulation and presentation of results (Darstellung). liese last
tic proceSses are obviously synthetic operations and are sometimes
grouped under that heading (Operations synthetiques)-

The preliminary bibliographical work is a much less formidable
task n01 than it was in the Liddle Ages when sources were widely scat—
tered, intercomnunication and travel, negligible library practice exceed-
ingly cnude and the cooperative facilities between libraries and other
repositories of historical knowledge practically non—existent. In that
day the search for and IOCation of documents was a very considerable part
of the historians task. Today in undertaking a historical project one
is aided by general historical bibliographies and numerous special ones
either in libraries or in secondary 'orhs. Libraries themselves have so
developed their systems of cataloging that the finding of sources is
comparatively easy.

The work of uncovering new sources is proceeding apace,and
together with the necessary critical Work of identification, constitutes
a Specialized field in itself imich the historian concerned with the
vriting of history is less likely to undertake than formerly. In fact,
in the fields of modern history'esPecially, the available sources are
frequently so numerous that the historians task is likely to be one of
selection and combination rather than of discovering new docunents
hitherto unknown to libraries and bibliographers.

It is recognized distinction betteen physical and historical
research that whereas the former employs the method of direct observation,
the latter must depend on evidence of the fact as reported by one or more
previous observers, either orally, in writing, 0r through material remains.
In the case of oral or written reports it is neceSSary to make allowance
for the warping or modifying effect of intervening psychological reactions.
The unreliability of individual processes of observation is a matter of
daily eXperience. Even when we are well disposed toward accurately regis-
tering the impressions which we shall transmit to others, both the ear
and the eye are exceedingly fallible, and the memory of impressions still
more so. Uhen this fallibility of the senses‘is intensified by pree
judice, self interest, hatred and other emotions it becomes literally

3" 343

true that ”the eye sees in an-object n at the eye brlrgs power of socing.*-
Censer ntly source material must be thoroughly tested in order
to determine th eruineness of the document; — the localization of the
document, — that s, then it originated, where it originated, end the
identity of the author; the value of the source, and the Validity of the
facts contained in the source, and the possibility oi supply gaps uhich
may appear in the body of facts contained in it.

1

p-

(L

3.8
g
i

The processes involved in reaching conclusions on these points
are essentially analytical and constitute that division of historical
method known as criticism. It has been customary to recognize tno sub—
divisions: namely, external criticism, dealing with the genuineness of-
the source, localization, and editing, and internal criticism which is
concerned with the determination of the value of the Source, the inter-
pretation of the source,and the establishment of the facts.** .It will
be noted that internal criticism does not deal exclusively with the
study of the contents of the source, for some of this analysis is re—
quisite in determining genuineness and localization.

The historical critic, sepecially for the work of external
criticism, is likely to enlist the aid of auXilizry sciences, such as
paleography, epigraphy, philologyythe knotledge of the peculiarities
of languages at particular periods, diplomatics, lexicografihy, technical
chronology and sphragistics, and-a lnotledge of the historv of literature.

The first task of external criticism is to detennine the gen-
uineness of a.document, for the historical pethter is strewn with instances
of historical forgeries. The first process is to consider the writing,
language, style and composition. In this task the auxiliary sciences of
paleogrephy and philolOgy are very essential, particularly for early his-
torical materials. These sciences have been so well developed that
ffibgery can frequently be detected without going further, If these
tests do not prove aquuate, the critic may next consider the contents
of the document from thestandpoints of their agreement with other contem-
porary documents, whether the Writer exhibits ignorance of things with
which a contemporary should have been faniliar or knowledge of conditions
and events he could not have known at time of writing. Finally, the
critic may apply the test of the agreement of the contents with the
general course of historical evolution. The critic may also consider
the conditions surrounding the discovery of the document, the use by the
writer of documents which could not have been known to him at the time
the source is supposed to have been written, and the motives which may
have constituted a stimulus for committing a forgery.

 

 

* Cf. Fling, F.I., Outline of Historical Method (1899) p. 13

“This classification is that of Fling, 'Outline' of Historical
nethod. It represents an attempt to compromise the class-
ification of Bernheim and of Langtons and Seignobos.

344

rhe usefulness of a document depends not only on its3éenfiineness
but also on the authorship and the time and place of its conpostiion, which
are not Lecessarily identical with name, place and date on the title page
or inscription, eSpecially in the case of the earlier sources. It has
already been noted that some of these facts of localisation are of imu
portance in determining genuineness, and in general the determination of
' genuineneSS and of localization are interrelated activities and somewhat
the same kinds of critical methods are requisite.

Finally, not all parts of a given source are of equal value.
Some parts may be the reporting of a direct observation, other parts
may be hearsay. In some parts the author may have been influenced by
prejudice, in others he may have had no motives for not reporting the
truth as he observed it. In some parts the author‘s facilities for ob-
serVation or ascertaining facts may have been adequate; for other parts
inadequate. Some of the facts may constitute a technical field with
which the author was not familiar, and therefore one in which his state-
ments are likely to be inaccurate. It has been well said that "failure
to analyze_their sources and to distingrish between What the witness
knOWS first—hand and what he has derived from others is one of the char—
acteristics of the uncritical historian."* In this connection it is
particularly desirable to analyze the relationship of various sources
to one another in order to determine their dependence or independencey
It follors from this that merely the number of references cited in sup—
port of a statement is not Conclusive unless the citations are Kmtually
independent. Lack of independence may not be due merely to the fact
that one author deliberately cepied from another, for both may have ob—
tained their facts from a common source or may have been affected by the
same set of conditions in making and reporting an observation.

Various methods are employed to test the matter of independence.
If the two authors see a condition or event exactly alike and follow the
same order in reporting it, this constitutes strong grounds for suSpicion,
for such identity of View is exceedingly improbable. The inCOrrect use
of the same expression by txo different authors is another ground for
suSpecting dependence. The question of thich of two related sources is
the original and which the derivative may also be established in Various
Ways; for instance, by finding an expression corrected in one manuscript
Which had been used incorrectly in another, noting a style that is free
and flowing as contrasted with one that is rough and incoherent, or find-
ing an account modified from another along the line of the known pre-
judices of the author - these.are likely to be indicators of derivation.
By the comparison of related sources, it is also possible at times to
reconstruct one or more gaps occurring in a document.

As already noted, some of the processes of internal analysis
are neceSSary in attaining the objectives of external criticisn. find
some of the steps taken in reaching these objectives will also serve
in attaining the objectives of internal criticism, which consist in

 

*Fling, Outlines, etc. p. 49

_,._,

./‘

r" 34

L}!

determining the value of the source, the interpretation of the source,
and the establish ent of the facts. '

The value of a source is determined by consideration of its
general character, the individuality of the author, and the circumstances
of time and place. Daily ne~Spapers, trade Journals, technical journals,
works of travel, census enumerations, political pamphlets, executive
messages, reports of technical commissions, legislative hearings, depart—
mental reports, 'ndividual diaries, account books, and personal reminis-
cences are by heir very natures of widely different values as sources
of information, irresPective of the questions of authorship and circum-
stances of time aid place. Furthermore the Value of these classes of
sources depends very largely on the type of facts which one is seeking
to obtain from them. It is desirable to illustrate this at some length
with particular reference to the requirement of agricultural research,
recognizing, he ever, that dogmatic estimates of the value of such laré‘;e
classes are likely to be inadequate and cannot take the place of in—
dividual eXperience and judgment in the evaluation of sources.

It has been the experience of the present writer that secular
neWSpapers are not a very satisfactory source of agricultural data, But
this observation is subject to exceptions. In recent decades, large
city dailies present such a diversity of information that one might
find a Valuable article descriptive of a region or an industry. Further—
more, in an earlier period the price quotations of daily papers are
valuable, although until neél into the nineteenth century not many
papers had a systematic policy of printing price quotations. On the
other hand, in recent years trade papers often contain quotations which
are fuller and more reliable than those of many dailies.

So it is with other kinds of sources; they may be serviceable
for one claSS of facts and vorthless for another kind. A political
pamphlet might reflect the views of the party or candidate although
wholly unreliable in its statement of facts. ;gricultural journals
contain records of technological interests, but until recent years
comparatively little information on the economic status of the in—
dustry. Earlier agricultural journals were frequently composed
largely of articles copied from other journals, sometimes slightly
modified, and not infrequently without credit. Gorksof travel are
not highly regarded as sources for episodical material, which the
traveler obtained from hearsay but frequently are of the highest ValUB
as circumstantial accounts of Conditions and practices observed by the
author. Government documents are the ultimate source for trade sta—
tistics, but many government documents are secondary rather than prim-
ary sources, representing compilations of data from earlier sources.
Furthermore they are by no means infallible in reproducing earlier
data, particularly statistics. Occasionally an earlier figure is
changed in a later document for purpose of consistency or in the
light of subsequent information, but provision is not aIWays made
to enable the reader to determine whether the change constitutes a
mistake or a deliberate readjustment.

346

For a particular class of source, obviously much will depend
on the personality of the author. He may be cautious and critical in
Working and reporting observations or temperamentally reckless and
extravagant. He nay be constitutionally honest or dishonest. He may
or may not have sufficient technical experience to obsc o and report
accurately.

An individual's background has much to do mith the character
of his report. English trcrelers, for instance, accustomed to the
intensive and careful agriculture of their own country were inclined
to advocate such methods in the meal; settled regions of the United
States and to condemn all that they saw without appreciating the fun-
damental differcn es in economic conditions. Non-technical observers
may employ language in describing technical methods thich must be in-
terpreted with care. 'For instance, early English travelers and col-
onizers were accustomed to use the word "corn" to apply to small grain.
Study of the context is necessary to determine whether they meant wheat
or maize. Beans and peas were used interchangeably by many early '
writers. The French eXplorers on the lower Mississippi Spoke of oXen,
cows and calves among the Indians; they were talking of bison.

The same man also will observe and report the same fact
differently by reason of circumstances of time and place. At a par-
ticular time of life a given individual may have a different attitude
than he tould exhibit earlier or later in life. At one time he may
be swayed by prejudice or passion and at another time disinterested.
Furthermore, the individual's attitude at a given time is likely to
be affected by external conditions, as for instance, his financial,
social, or political relation to others, the general state of public
Opinion at the time of writing, and conventional attitudes of his

. . . . CIVWer nce; . . . . . .
assoc1ates. Sometimes also inconsis areas are such that an 1nd1v1dual
did not have opportunity to observe the facts. For instance, a
writer who undertook to describe the agriculture of a State may have
spent most of his time on the principal routes of travel.

. All theSe phases of criticism are of use not only in de-
termining the character of the source as.a whole, but also in es— ‘
tablishing the facts, which is the ifinal goal of historical criticsm.
This statement,-honever, is likely to involve not merely the study
of a single source, but of many. For one thing, a single source
usually does not contain all the facts needed for a particular his-
torical synthesis. Indeed, the number of requisite sources may
range from'one or two up to thousands, depending on the scope of'the
synthesis. But beyond all this, a number of sources may be required
for the establishment of a single fact, particularly as sources differ
in reliability. '

The general.rule is that a single reliable source may es-
tablish the existence or character of e.conception or of an attitude,
particularly if the attitude of the writer of the source itself is
the important fact to be ~stablished. Similar to this, are the ele-'
ments of reality which constitute a part of a Whole that may be im.

347

aui’sivnative. T r instance, the account of a battle_rhv he rfthical,
but the weapons sentionc' in the account, the socia relntionships
casually exhibited, and the superstitions and intellectual attitu es
reflected in the accornt may be separable as protable realities from
the imaginative xmole. Ehe affirmation of a single source concerning
an external fact is not considered surficicnt for the unquestionable
establishment of such a fact, though it may be excellent presumptive
evidence. Two or more creditable and independent sources are con-
sidered neceSSary for the establishment of an external fact. Two may
be enough, cepecially if the accorhtc agree on many details which have
a casual rather than a necessary or usual relationship to the fact.

If two or more son‘ces contradict one another, this contradiction

must obviously e reconciled either in terms of the lack of reliability,
biaS, lack of independence, or general ingrobability of the accounts
supporting one 01 the two positions.

 

 

Ihile a cautious and critical attitude1fi£>necessa‘y in hand—
ling all historical materials, it is obvious that the more r cent the
‘material the lees difficult and technical are the problems 0 external
'criticism, or for that matter or internal criticism. Coxpare for in-
stance, the difference in the nroblem or a historian dealing :ith sources
of the period from 1850 to 1860 with the task of a historian tho is trying
to ascertain the facts of the period from.1250 to 1260. For the latter
period his sources will be conparativel; ion, they will be reproduced by
hand rather than printed, and Lrohably in medieval Latin. They must be
carefully compared with other known doc ments to determine genuineness.

A minute knowledge of paleographf‘nocetsary‘to determine authorship: and

a comprehensive knowledge of the literature and conditions of the period
in order to determine place of corpoaition, and cirCtxstances affecting
the attitude of the author. in echrt knovlcdge of contemporary docua
ments m y be requisite to determine dependence or independence. The
question of the real date oi composition, as distinguished from the
nondnal date, may need to be determined by a nirnte applicatiOn of
the principles of philology.

1.
e
f

It is obvious that much'or this \ill be unnecessary in the
handling of recent historical data. "The technical apprenticeship is
relatively short and easy for those rho occupy themselves with modern
or contemporary history, long and laborious for those who occupy them_
selves with ancient and medieval history."* Sometimes there is even
a tendency of historians to modify the relative importance of criticism
as compared with other methods of historical research. '"It is quite
possible, Whatever may be said, to have the historical sense in full
measure without having even, both literally and figuratively, wiped
away the dust from original documents.- that is, without having dis-
covered and restored them for oneself."** In fact the synthetical
historian may well afford to leave the tasks of external criticism to

 

* Langlois, Charles V. and Seignobos, Charles, Introduction to His-
torical Study. Translated by Berry (London. 1912) p. 55.
** Ibid., p. 114 '

348

st in V“" ' ' ”" of We:

the specialist in that field. The two types of work frequently re—
quire different temperaments, aptitudes, and training not often com-
bined in the same individual. Critical scholarship for instance,

”may come to lack perspective and be characterized by excessive pre-
occupation with little things.“* This may lead to hypercriticism,
fear of publishing perfectly valid data, and a tendency to lose sight
of the large historical objectives. The synthetical historian, es—
pecially if he be dealing pith a recent period, is likely to find the
genuineness and localization of his sources already determined in the
caSe of the great majority of Cases, and therefore can Spare himself
much of the drudgery of external criticism; he canrnt avoid the ob—
ligation of concerning himself with such critical tasks as determining
dependence or independence, the author!s facility for making observa—
tions, absence of bias, the ability of the author to observe and report
accurately, and the relation of various sources to the establishment of
the facts.

Lhile a long step forward has been taken when historical facts
are establiShed, there yet remains the important task of determining the
meaning of the facts, that is, their relationship to one another. This
task, together with the presentation of the facts is the field of his—
torical synthesis.*t

Relationships, of course, may themselves be questions of
fact. For instance, the occurrence of a great flood in a given year
may be established as a fact. The subseouent rise of price of a given
product grown in that area is another fact. The reSponsibility of the
flood for the rise of price is a relationship which may be established
as a fact, This might involve the assenbling of many other facts,
and the establishment of other relationships negative or positive.
Students of price analysis will appreciate how many other factors
might have to be allowed for such as total Size of crop, other factors
affecting it, carry—over; demand conditions, etc., before the influence
of the flood on price could be determined, and particularly before the
extent of influence may be measured.

In the great majority of cases, however, relationships can-
not be incontestably proved - that is, established as facts. At best,
we can determine a probable qualitative relationship, and an argument
may be built up to support this probability. Quantitative relation—
ships may be established having a greater degree of probability pro-
vided the raw materials are available on all the different factors
that must be alloWed for in establishing relationships. However,
the farther we go into the past the more it is likely to be necessary

 

* Ibid., p. 129 ,

** It is obvious that that the historians call synthesis includes
a combination of analysis, as this terms is used elseuhere in
the handbook, and synthesis.

349

to depend on inference from partially knovn facts. It is most import—
ant that the historical student discriminate between definitely proved
and only partly proved relationships and particularly mere congectures
and working hypotheses. Failure to be thus discriminating has re—
sulted in much bad, though possibly, entertaining history.

Methods of logic, of course, are available as a means of
testing the proceSs of reasoning, but then quantitative measurements
are not applicable the relative weight and inportance of Various facts
in a relationship must depend largely on opinion. This fact, as well
as the continual discove y of new materials, is responsible for the
necessity of continually reqriting histomy.

There are almost innumera 1e kinds of historical syntheses.
Many of the earliest were mere chronologies relating events in the
order of occurence. Evan in these crude chronologies there was ev—
idently a very large element of selection since all the experiences
and happenings of human life were not included. As already indicated,
many of the earlier syntheses constituted selections of important milia
tary and political events. Sometimes a dynasty or a prominent individual
Was the canter'of interest.

, A very comnon synthetic unit even in modern historiCal writing
is gBOgraphic, as, for instance, g,nation, a State, a county or a city.
Usually this geographic unit may he further defined by limiting the
synthesis to a period. The synthesis may include many tOpical phases

of the geographic unit and period selected, such as political, military,
diplomatic, econoadc, cientific, literary, and social history, or it
may be limited to one phase, as for instance, economic life, or to a
still narrower topic, such as the economic history of agriculture, or
even of a partiCular crop or kind of livestock.

It follows from this that a synthesis may be wide or narrow
in scope. It may be world—wide (Zeitgeschiche), or nation—wide, or
it may be limited to a toxnship; it may conprise a brief period or a
long one; it may be limited to a single phase of hunan life or com—
prise as many phases as the historian.considers significant. Both
types have their place. The broader and longer syntheses permit the
determination of the wider relationships and sequences. The syntheses
of narrower scope make possible the consideration of many facts and the
determination of many of the more minute relationships that would
necessarily be omitted in the former type.

The fiselection of a topic will depend not only on the in—
terest but also on the eXperience and aptitudes of the worker, for
the broader syntheses, other things equal, are the most difficult.
It is well also to consider carefully whether the subject needs in—
vestigation because it has not hitherto been investigated or in-
vestigated inadequately, because new facts hays become available,
or perhaps because it appears that a new and more illuminating his—
torical synthesis may be achieved.

._..I --—~‘_.‘....

1 ’ I.
, t
35C '

It is obvious that the nature of the historical problem and
the methods employed zill differ greatly in_accordance ride the scope
of the synthesis. For instance, there will'be a difference in the
completeness with which available sources are consulted and in the
extent of dependence on secondary sources. No man in a lifetime could
consult all the sources of American history. It is doubtful if he
could consult all the available Sources for even a limited period such
as the quarter of a century following the Civil War. For such a
period it is a question not of consulting all the sources, for in a"
sense every private letter, every pamphlet, every newspaper, and every
government.document is a potential source of more or less significance.
Clearly the problem that presents itself is one of selection, not only
of the worththile sources, but also of the worthwhile facts within the
sources.

Ehxthermore, in the case of the wider syntheses the historian
will be compelled to rely to a considerable extent on sound secondary
sources, particularly scholarly monographs of limited scepe. It is
true, no historian ean feel perfectly safe in adopting the results of
another's work,."but so complex a science as history, where facts must
ordinarily be acounulatcd Ly the_million before it is possible to form-
ulate conclusions, cannot be built up on this principle of continually
beginning afresh. In order that science may advanbe it is necessary
to combine mhe results of thousands of detail researches."* It has
been well said that we should be carefu ‘to avoid the fault of those
who"heed only the unprinted aid scorn the rest, resembling those
tourists who have no cease until they have got access to some private
gallery, but merely glance at the public ones; there the best pictures
may happen to be.** Nevertheless, he who employs secondary sources
is under even greater obligation to apply critical methods than in
the case of primary sources. I

In every historical synthesis there are certain facts that
are required as background; they do not.constitute Lhat may be called
the "contribution” or essential core of the synthesis. For such
facts particularly, it ray not be necessary to delve into Original
sources if sound secondary sources are available. '

The more essential problems of historical synthesis, have
been summarized as follou::*t* (l) to set bounds to the subject;
(2) to-divide it into periods; £3) to decide what facts are to form
part of the synthesis.and that are to be discarded, (4) to_determine
- what causal connection exists between the parts (5) to decide What
change has been effected by the historic action (6) to determine "hat

 

* Langtonls and Seignobos, Introduction, etc. pave 280, Cf. also
filing, F.l. .The uriting'of History, page 88. ' ’

** Jusserand in The Zriting of History, p. 19

*** Fling, 3.15., The 'Cfriting of History, p. .127

.351

parts shall be emphasized and what touched upon but lightly, (7) to
determine hon many data shall be used for the sake of color to pro-
duce terisimilitude in the reproduction of the past.

The scope of this article does not admit of a detailed dis-
cussion of these and other problems of historical synthesis. But it
would not be proper to close this~discussion without a turd on that
very essential phase of synthesis, namely, the adequate exposition of

results; There is a perennial conflict of Opinion as to the extent
to which historical writing can be sound and still be so attractive as
to be widely read. The growing unpoPularity of history as literature

for general readers recently led the inerican Historical Association to
appoint a committee to inquire into the causes of the tendency. In
its report the coundttee todk an optimistic view of the possibilities
of improvement in this regard, and the Chairman, Ambassador Jusserand,
expressed the View that "an historian who uses so dull a style that he
will not be read is as useless as a painter Who should use invisible
colors. He is, moreover, sure not to do justice to realities, thus
swerving from truths, for realities are not dull."*

There is obviously no reason for slovenly English or crude
methods of expression in presenting the results of historical investiga-
tion. It does not follow, however, that confermity to high standards
in these regards, While still obserVing scientific standards, will re-
sult in a "popular” product. It is well to have popular histories
provided they are accurate, but to present adequately the results of
original investigation necessitates exhibiting for critical examination
by others of the essential scaffolding on which the conclusions rest.
This involves a generous use of footnotes, and the inclusion in the
text of many facts which, introduced for purposes of confirmation, may
prove dull reading for all but critical students concerned with'the
establishment of historical truth. Professor Fling has well Said,

“It is no more reasonable to eXpect that all historical work should be
written for the general reader than that all works on natural science
should be accessible to the same class of readers."**

 

* Jusserand and others, The Writing of History, page 5
** The Writing'of History, p; 158 ‘

352
‘ BART EQUR

THE PRESENTATION AND UTILIZATION OF RESULTS.

 

A. The Use of Tables for Presenting Results.

Well set-up tables are an effective measure of presenting
large masses of information in small space. If properly constructed,
they reveal on their faces the relationships which have been developed
by the analysis. A well designed table will convey for more informa-
tiOn than can be given in an equal Space of text. Tables are, in
addition, the only satisfactory way of presenting the details of data
whichfall naturally or otherwise into classes of pretty much the same
kind of. thing. _ y

The 'form and Construction of a table depends in part upon
the purpose it is to serve. Tables may roughly be clasaified into:

(a) Storage tables,'in which large masses of date. are arranged
in convenient form for reference or to be worked from.
The large tables in the body of the census volumes are
examples of such use of tables; also the tables in ep-
pendices of bulletins.

(1)) Analytical or demonstration tables, whose purpose is to
reveal relationships that have been deve10ped and
provide the principal supporting data. Most tables
in the main body of bulletins, and the tables in the
introductory portions of the census volumes, are
examples of this use of tables.

(c) Display tables, serving very nearly the same purpose
as cartogrems.

The two latter types of tables have a definite story to
tell, the demonstration table emphasizing the clearness and accuracy
of the presentation of it, the display tables emphasizing the effect-
iveness and attention—Value of the presentation. The titles and
headings of both should effectively suggest the story of the table,
of demonstration tables, the exact content in addition.

The textbooks in statistics nearly all present the technique
of table-making. It has therefore not beondeemed necessary to dis-
cuss the subalect systematically in the handbook. Instead, Mr. S. W.
Mendum, iho for five long years has been editing tables prepared in the
Bureau of Agriculture Economics, has been asked to summarize his con-
clusions as to procedure and practice in preparing tables for publica-
tion. His discussion refers to storage and demonstration tables, not
to display tables.

353

Preparation of Tables for Publication.*

General V Tables should tell rather definite
comments. stories. The story to be told dictates

the form and Complexity of the table,
and the relation of the tabular matter to the‘hext in some measure
affects the minor details.
tables

The form of the statisticalAis more or less constricted
by the information available, the anticipated needs of users and
the size of the medium of publication. Tables should be as
simple as possible, but simplicity is often very illogically de-
fined. Only so much information should be thrown into a table as
is definitely related to the point to be made. It is a mistake to
load anything into a table because there is space on the sheet for
more lines or more columns, and it is still more of a mistake to
leave out data essential to proper understanding of the story the
table is sXpected to tell. If a table has Very few figures, it
probably would be best to cast it into text.

The simplicity of a table is more clearly measured by
the care taken in its preparation than in the aflbunt of white space
it Shows. The blackest looking table may be of the simplest type;
because it tells its story without possibility of confusion. Thus
the worst appearing tables in the Yearbook of Agriculture are the
tables of temperature(and like tables); yet there is none more
simple. Many another relatively open table is much more complex.

The two greatest weaknOSSes of tables as submitted for
publication are their order of arrangement of the data and their
use of symbols and abbreviations. The most natural and convenient
procedure in assembling the data needed for a table may not be all
satisfactory for the presentation. It is too much to eXpect that
the reader will be versed enough in the matter presented to be able
to interpret exactly and promptly all the abbreviations and short—
cut eXpressions and other symbols an author may find it convenient
to use on his work—sheets. These have a habit of persisting in
the printed publication.

Procedure in No definite statements can be made
Table Construction as to the best procedure always to follow

in setting up a table; but there are a
number, which if used with discretion, will help considerably:

1. Plan the tables for the specific purpose in hand.
Do not expect to use the same arrangements as on the
work-sheets. The nature of research is such that the

 

*By S. E. Mendum, Bureau of Agriculuiral Economics.

1

4.

5.

order of doing Wonk is often quite different from the most
effective order of presentation.

Pay more attention to the relationships of the matter in
a table than to the form and size. Later on some one will
raise the troublesome questions incidental to printing
large—size tables. In general the printer can and will
print soccptably any table requested.

Test each figure.‘ Tables are made up of columns and lines
duly labelled. ~The reader has the right to expect and
assume that any figure in the table is correctly described
by both the boxheading at the top of the column in Which
it stands, and the item in the reading column, or stub,

of the line on which it stands. Figures that do not meet
this test exactly should be individually further described
or limited by footnotes. If the footnotes become too
numerous for Convenience, it is likely that the sorting
has not been so close as would be desirable and more lines

' or more celumns, perhaps more tables, are indicated.

If this test is applied figure by figure, con-
scientously, fundamental faults in tabular matter will be
avoided in the.early stages of the work. The wisdom of
the_groupings is likely to be challenged by the first
half—dozen figures tabulated, or if the groupings de-

cided on are correct, the descriptions may need to be
changed. The accuracy of corp uted or derived figures
must be unquestionable.

Test the symbols and abbreviations. The use of unfamiliar
symbols is confusing ~ the bad effect on rec.ders often off—
sets the Savings achieved through condensation. There are,
however, many‘practices which may well be standardized so
that they will have the effect of faunliar symbols and thus
relieve tabular matter of all unnecessary effort on the
reader‘s part.

Do not try to make your table so all—comprehensive that
one table will serve a dozen purposes. Such tables will
satisfy no one in particular and will draw more comments
unfavorable to your judgment than they will excite praise
for your cleverness.

Think twice before you use a table after it is made.’ This
is especially true of the most simple and the most complex.
If the main object is illustrativenor argumentative, it is
quite possible that a text statement will serve much better.

' 355

EWQ As in a graph, the independent variable
EEQQEngfi. is always plotted with respect to the hori-

zontal or x-axis, the dependent variable with
reSpect to the vertical or y—axis, so in tables
' the results are most satisfactory when the independent variables, the
items described or measured, are listed in the reading column or stub,

and the description, basis of measurement, or characteristic on which
sorting is made, is placed in the column headings and title. long
narrow tables (one or two columns of figures) can be doubled up for
economical printing, and wide tables (of few lines) can be broken over
without loss of effectiveness, the stub being repeated.

There is somewhat more advantage in uniformity than appears
at first glance. Individual tables may be as serviceable built hori—j
zontally as though built vertically, but miscellaneous shifting of fonm
in a publication without abundant reason is a weakness that should be
avoided.

In a publication containing numbered tables, it is usually
best not to try to insert unnumbered or "text" tables9 Data for such
tables may be presented as text or as part of an appropriate numbered
table to Which reference may be made.

The adequacy of the title to the table, or the difficulty
of preparing such a title, are serviceable h;diCators of faults in
construction of tables, for the title should describe the table comp
pletely and accurately, and if, together with the column and stub-
headings, it does not do so readily, probably the table needs re—
planning.

Titles descrve more attention than they usually get from
authors or editors. Some authors seem to be Satisfied with a
general topical title. Titles should be brief, but the additional
requirement that they be also clear, precludes general use of the
topical title alone.

The key word or phrase followed by a colon makes a good
start. Thenthe specifications of the table can be enumerated as
far as necessary. Such titles may seem unnecessarily stiff and
formal at times, especially when there is a long series of tables
on the same subject. Judgment and moderation is needed in this as
in all things.

Where the general subject is mechanically subdivided by
sorting and grouping, the attention and interest of the reader be-
comes directed more toward one of the specifications or factors in
the table than to the general subject. Thus, in a publiCation
dealing with crop production in the South, the titles to tables
dealing with cotton alone might well begin with the woni”cotton"

356

as "Cotton ; acreage under cultiVation July 1, by states, 1913-1926."
If all ten cotton tables were assembledin'one section, however, the
title to the same table might properly read "Acreage "of cotton under
cultivation July 1, by states 1913—1926." The more complicated the
tables must be, the more urgent the formal plan becomes, and then
uniformity of appearance dictates that the simple table also be cast
in similar mold.

Avoid subtitles as far as possible. If the informtion is
necessary, it should be in the title; if not, it should be cut out,
as it is likely to be lost anyhow.

The words "weekly" "monthly" and "yearly" can often be
dispensed with to advantage.

Use of capital letters to indicate importance of the-nerd
or phrase capitalized as a substitute for care in writing a title,'is
a confession of weakness. '

Dates indiCating periods covered should usually be placed at
the end of the ti tle.

For printer's convenience,‘ tables occupying more than one »
sheet should bear the notation "Continued" in the lower ’righthand
corner, and again at the end of the title at the top of the suceed-
ing sheets.

Dates, especially tho se designating periods covered by
averages and usually hyphenated, are the bane of the careful stap-
tistician. He can not tell whether they are inclusive or no t.
Uniformity of practice is more important than any given practice.
The practice which will help most is a general agreement that the .
hyphen or "to" between dates shall alvays mean that the dates shown
are included. Thus "190294.913" should be used as covering the
five full Calendar years beginning Jan. 1, 1909, to and including
Dec. 31, 1913. An average ‘made up of figures beginning as of
July 1 should be identified as fiscal years, or as "Years beginning
July 1," or as "Years ending June 30", or "July, 1909 —‘ June 1913.”
An average of five crep-years beginning July 1, 1909 would be desig-
nated as the "Five—year average, 1909-40 to 1913—14" using the word
"to", still in the inclusive sense,1n ' inpreferen ce to the hyphen.
The form "1909-13" Should be discarded for a five-year p eriod.

Every table should have a date or dates somewhere.

Size groupings in the saris series should also be always
mutually exclusive.

' 3b?

Reading columns or stubs usually give little trouble. The
items must be definite concepts or the table trill be confused and
obscure in use. ' Sore classification my be given effect by pro-
gressive indentations; each subsequent class indented the same dis-
tense) and by center headings in one or more sizes of types. Hor-
izontal rules do not cross reading columis in the best practice.
(At times two or more tables with the same boxhoadings may be 8.3--
sembled on the sane pge Without repeating boxes vvhen‘space- is a
consideration. ) » '

Reading columns may or may not include the units to nhich
the figure to the right apply. Such inclusion is particularly
serviceable when the table shch prices, costs, or values" in dollars
for a. variety of items. Even then,‘ a Column showing designated.
units is prefera'ole.. '

Box headings are the authors means of indicating relation-
ships. As thetable is read from‘lef to right, the earliest dates
in time series should be on the left., Distributions are indicated
by subdivision of the box from top to bottom. Three or four sub-
grouping is about the-limit before the question of‘dif‘ferent tables
or differentrstatement'of the title arises. ' I i l

- Box headings must usially be short'but they mist be def-
inite orrade definite by (annotation. ,' The reader expects. to, find
in the column Only those items‘rfnich are accurately described by
the contents of the boxheegdings above it.

Readers do not refer to the box head, rating it a part
of the reading of the figures, often enough, There is‘ground
for the opinion that they have found the practice of so little
help that they have got out of the habit.

A practice'desirable on the whole, though oniCWard at
times, is the placing of the wait deSignation above the..figures
in each and every column. The box heads are relieved of the
burden of carrying the unit; but the chief advantage is that with-
the unit. so placed. there never can be any doubt as to the unit.
It serves the writer as a constant warnin not to mix units.

When the units are all dollgrg or all agree. or counts it is'h'ard
to see why the unit designation is needcdother than in the title.
Uniform practice is the only reason for'it. Such words usually '
do not often'teke Space needed. The conventional dollar sign
($) does, of course. Units should be printed in italics.

Punctuation should be-limited to' the essential.

Box head desigiations-fOr marking columns should be in
the singular number. - v ,

558

Totals and averages logically appear at the bottom or
right of the detail which they summarize. Sometimes totals are
placed at the top of the figure columns, or sections of it, but the
net gain is negligible.

Totals are placed before average.

In a summary line containing totals and averages, the situ»
ation is amply covered by the use of ”tetal" only in the reading
column, eSpecially in those cases where the box above the average
indicates a column devoted to averages.

There is disagreement in the matter of checking tables,
especially tables including derived figures or series of related
tables. The scientist is tempted to record the individual item
correctly and let the total come out as it may. The practical man
raises the question of how he is to tell ”what is what" when dis~
crepancies are apparent. Tto total figures of the Same description
should be identical numbers even if twenty tables apart. The total
percentage must theoretically be 100.0 and the detail must add to
that figure. Averages must be obtainable by simple division of
items shown or definite warning must be given by statement as to
how they have been computed. Totals of rounded figures must be
the sum of the detail as shown. An average percentage is computed
from averages of totals, pet from the total of percentages. Averages
should not be worked from incomplete series — a yearly average should
not be worked from 11 months' data. Ayerages of derived figures
should be computed in the same manner as the derived figures thus
summarized.

Fbr American readers, foreign money (prices and Values)
should be converted into domestic equivalents.

Footnotes permit much wider use of some material of doubtful
character than would otherwise be preper, and they prevent many mis—
takes of interpretation.

Fbotnotes should be used as liberally as needed. Numbers
to refer to footnotes rather than symbols or letters are practically
necessary for all but the most simple tables. Footnotes run con»
secutively from left to right and down the page.

An unnumbered footnote giving the source and other general
infbrmation about the table is often desirable. Such should appear
at the bottom but before numbered footnotes.

Columns of figures subject to regular revision may be in—
dicated by the word "preliminary" in the box—head or by footnote.
Individual figures in a column questionable on any score should be
covered by a specific footnote.

359

Totals and averages are usually to be footnoted if and
whenever any of the component figures are definitely footnoted.

If data in a table are not quoted as given in the sources,
state electly how they were obtained.

Numbered footnotes should be used for sources of figures
in several different columns of a table. Sources should be so
definitely described that one using the date. may retrace every
step taken. Every citation should be so specific that the par-
ticular data may be readily identified. A blanket form of reference
is permissible only when the table cited appears in the same form
continuously from year to year.

Arrangement of items is often troublesome. The alpha,—
betic order is always safe and definite. For statistics by states,
the order used by the Bureau of the Census has much to recommend it,
whether in tables of all the states or for any large section. Order
of importance on any single factor is a poor makeshift for careful
text statement,vand gives grief more eften than it points an argu-
ment. Any arrangement by size is likely to be upset at another
period. Arrangements adopted should usually be maintained in a
publication.

Use a zero in tables if and when zero describes the known
situation. When there are no data, use blank space or "leaders".

A. hand book of rules prepared for guidance of clerks in
the Division of StatistiCal and Historical Research and largely used
in the Bureau. of Agricultural Economics may be obtained from the
Bureau.

sec 7
aneurysm"

Graphic presentation, in the portrayal of statistics and in
statistical analysis, has become an important instrument in the field
of economics and business, in recent years. When properly used, it
is a means of so digesting and interpreting masses of statistical
material as to bring out features‘of data. that might otherwise go
unnoticed. The mind does not hold bare figures easily and relation-
ships between many figures are not easily and quickly discovered.
Graphic presentation tends to make data attractive and more easily
interpreted and therefOre more instructive.

. Graphic presentation is used in analysis. It usually fol-
lows tabular presentation and arithmetic computation. As it is used
as a means of clarifying date, it should present them in such a way
that the important points are easily discerned by the casual or average
reader.

Graphic presentation is not a substitute for accurate data.
A chart cannot be more accurate than the data from which it is nude.
In fact, untruthful charts can be made from true data if wrong methods
are used in portrayal.

' In the following discussion, it is assumed that the reader
is familiar with the general principles of graphics vfnich are found
in the books dealing with the subject. It is hoped here merely to
emphasize a. few of the principles vital to good presentation, to show
how these principles can be applied to Specific cases, and to offer a
few guides to the maker of charts and graphs in order that he can mice
his work as useful as possible with a. minimum of time'and expense. A6
costs of good graphs and charts, and even of their reproduction, are
rather high, the resulting prints in a publication must be of the
greatest use and value to justify the expenditure.

The fundamental principle in all graphicypresentation is
simplicity. A, complicated or confusing chart will defeat its own
purpose and my actually be misleading. The eye can evaluate to an
accurate degree only a very few fee-Lines of a chart at a. time. The
design of a chart should take into account the limitations of the
eye. Of the three methods of po rtraying Values-«ml ength, area, and.
volmne-u-length is by far the most easily gauged by the eye. All
values should be shown by this one dimension if possible. Area, in—
volving two dimensions, is evaluated much less accurately by the eye;
and volume, involving three dimensions, should not be used. Angles
and ratios are ordinarily evaluated less accurately than is length,
but they are moderately reliable. ‘

 

*By P.. G . Uainsworth and The'v D. Johnson, Bureau of Agricultural Economics.

361’
Selection 9f Illustrative fiaterial

The author of a publication should select the illustrative
material with a view to portraying only the main points of his dis-
cussion. It is possible to over-illustrate. Unnecessary maps,
graphs, and photographs are a decided hindrance to the effectiveness
of those which are esaential. If too many illustrations are used the
continuity of the reading matter is destroyed and the expense becomes
great.

A complicated or involved table should not be selected for
illustration, for even if a graph could-be made to portray the data
it would be just as difficult to understand as would the original
table. The average reader will not study an illustration unless the
main points are discernible at a glance. Simple tabulations and re—.
lationships make simple charts. This principle applies to photographs
as well as graphs and charts. An illustration which requires much ex—
planation to be understood can well be eliminated.

It is well to make a small layout of a proposed chart or map
to determine whether the data and method used-show exactly the points
to be brought out. The layout may well be subjected to the criticism
of a disinterested person because a disinterested reader will eventually
read and evaluate the illustration in the final publication. The
author who has a complete knowledge of all phases of the subject matter
may draw conclusions from.an illustration which are not evident to the
reader. If any doubt exists, the illustration can be well discarded .

Line and Bar Charts

Time Series
{‘4‘ '
The most important line on a line chart is the one that repre-
sents the data. All other necessary lines should be in the background,
to guide the reader in his interpretation. Line identifications, grid
lines, and all lettering, should appear to the eye as of secondary im—
portance (see figures 1, 2 and 3).

Time series line and bar charts are usually plotted with the
time element, the independent variable, running horizontally from left
to right. Short series bar charts are an exception.

When space does not permit the indication of each time unit,
alternate units, or every fifth or tenth unit, can be shown. A maze
of grid lines is a distinct handicap to the effectiveness of a chart.
The fewer the grid lines on a chart, consistent with adequate inter-
pretation of a line or bar chart, the more effective the chart.

The relative importance of 0 Curve in a line chart should be
in accordance with the importance of the data from which it is made.

'362

Usually a Solid line indicates the most important data; other designs
of lines are used for other data.

The curves in a line chart should be limited to those that
can be easily and accurately followed. A maze of lines is merely con—
fusing. If the purpose of the chart is to show the same general trend
of many series of data, all the series may be put in one chart, with
the thought that the trend is of importanCe and the variations of the
individual series are merely secondary. As a general rule, if there
are more than two lines in a.line chart, they should appear on dif-
ferent areas of the grid. If they cross and recross they become con-
fusing. '

. All data in a line chart are plotted in direct relation to a
base line (ratio charts excepted). This baSe line is next to the data
line in importance and should be so indicated on the chart. To discard
a base line on a chart may lead to distortion of the data and to false
conclusions except in instances in which the distance between the zero

- base and the curves is not the significant factor in the portrayal

(as in trend lines, etc.). In such a case the zero base can be omit-
ted and the bottom of the grid indicated uy a light uneven line, below
the line that clears the lowest point of the curve.

-5pace alloted to a time interVal should be uniform throughout
the chart unless otherwise indicated definitely by a broken grid.

When plotting two or more series of data having different
units (such as dollars and bushels), effort should be made to use scale
units which will give curve variations of somewhat the same magnitude.
True relationship can be easily hidden if distorted scale intervals are
used. If a large scale interVal is used for one series and a small in-
terval is used for another series, the chart will tend to distort the
data.

Lettering in a Vertical position on any chart should be
avoided. Legends for curves should be so placed as not to interfere
with the visualization of the curve.

The base line in a bar chart is even more important than in
a line chart and should be made more'emphatic in the bar chart. A bar
chart Visualizes the actual magnitude of a Value or‘quantity by the
length of a bar; except in very rare cases, the base line by which the
length is measured should always be made wide enough to emphasize it.

The nunber of values or quantities to be plotted within a
given space determines the width of bars to be used. Usuall y the best“
'appearance is obtained by having the bars and the spaces between the
bars of_equal width.

368"

Solid, hatched, or shaded bars are more effective than hol—
low bars.

General

These principles, applying to line and bar charts showing
time series, are generally applicable to all other bar and line charts.

As a rule, bar charts of data other'than time series are made
with horizontal bars to conserve space and give room for lettering the
bar designations.

Lettering or numbering in solid bars should be avoided if pos-
sible, and data figures should never be placed at the end of the bar.
If necessary, they can be put at the beginning of the bar outside of
the grid.

If bars are segmented, either on a percentage or an actual
value basis, segments may be distinguished by hatched designs. If the
Segments range from most important to least important, density shadings
with black the most important Can be used. If segments are of the same
importance, areal distinctions (see figure lg) through the use of various
designs of dots and lines which give all segments about the same density
to the eye, can be used. The smaller the number of segments, the more
effective will be the visualization.

When a series of data Can be shown either by a line or by a
bar chart, the bar chart is usually the more effective. Bar charts are
necessarily simple in construction and give a complete visualization
of actual data.

Ratio Charts

Ratio charts are used chiefly to show the relative or per-
centage changes in data. Equal upWard or downward slopes of lines on
a ratio chart indicate equal rates of increase or decrease (or equal
percentage changes) in the data (see figure 4).

A ratio chart is made on semi—logarithmic paper, which is
ruled with a logarithmic scale on the vertical axis, but with an
ordinary arithmetic scale for horizontal. All the standard rules
regarding the making of line charts apply to ratio charts with the
exception that no zero base line appears on a ratio chart.

Ratio charts do not show actual changes in Value of data.
They will he very misleading to the average reader if this limitation
is not well brought out in the discussion of the chart. Ratio charts
are excellent for the purpose for which they were devised, but should
be used with caution, as they are easily misinterpreted by the average
reader.

$364

If neither ratio chart nor semi-logarithmic paper is available,
the vertical scale can be laid off in the following says:

.1. By using divisions on any 3 tandard calculating slide-rule.
These divisions are on .a logarithmic scale. ‘

_ By using an ordinary ruler on which the inches are divided

in ,'uenths.' Let zero on the ruler equal the value of 10 or log 1,000

m: 10 inches on the ruler equal the Value of 100 or log 2-,000. Using
the first figure in the mantissa of the log as inches with the other
figures as fractions of inches, the logarithmic scale can be laid off
accurately. ‘ For example, the mantissa of the log of 10 (or 100,1000,
10,000 etc.) is zero. The mantissa of the log of 20 (or 200, 2000,
20,000 eta.) is .301.‘ Using the first figure of the mantissa (3) as
inches, and the remaining figures (01) as fractions of inches, the dis-
tance from zero on the ruler or the value of 10 (or log 1.000) to the
Value of 20 (or log]. 301) is 3.01 inches. ‘ n the same way, the distance
on the rulerI' fromzeIro to log 50 is 6.98 inches and to 90 is 9. 54 inshes.
The following Ilogariths are the only figures necessary:.

' I W I Log ari thm

10 1.000
20 I t 1.301
'-I_zo.: .~ a, 1.4%7

40 1.602.
50 1- .? ,,- 1.698
60 - 1.778

70 I ' ~ I l7 1.845.
80' ' 1.903

90 N := 1.954;

. 100 , 'I'V 0J7-12.000~

The above may be repeated for any increase needed in the scale
to fit the magnitude of the data, by increasing the characteristic of
the log. The range between the characteristic of the log of the lowest
value to the characteristic of the log of the highest value in the data
gives the nurmer of decks which the chart requires. For example, if the
data. ranged from 100 to 10,000, the _characteristic of lOO is 2 and of 10,000
is 4, and their difference is 2. Two decks would therefore be necessary
to plot the data. '

After the vertical scale is laid out and grid lines are drawn,
data are plotted as in ordinary line chart.

By converting original data into logarithms and plotting these
IOgarithms on an ordinary arithmetic scale, the same shaped curve will
result as When the original data is plotted on a logarithmic scale.

This method, however, is used only in rare cases in technical statistical
analysis and has no place in publications of' general character.

1113 Diaggarns

Pie diagrams or componentepart diagrams are used to present a
number of individual items in relation to their sum (see figure 5).

~ In constructing a pie diagram the data should first be expressed
in percentage form and the Various segments should then be laid out With
the help of a percentage protractor or degree protraetor. Plotting of
values should begin at the top of the circle.

Density shadings can be~applied to the segments of a pie diagram
to bring out their relative importance. '

Proportionate sizes of two or more values can be shown by variap
tions in the areas of individual pie diagrams, remembering that areas of
circles vary according to the squares of their diameters.

Pie diagrams are less effective than bar charts in portraying
nest data because the eye measures length more easily than angular widths.

y
Scatter piagrams

Scatter diagrams, also called correlation charts, or two-way
frequency charts, are used as an analytical device as Well as a method
of graphic presentation of the relationship between tee variables. The
dependent Variable is generally plotted on the vertical Scale and the
independent on the horizontal. The position of the dot representing
any observation is determined by reference to its'numerical Value on
the two scales at the intersection of the two values projected. The
scale should provide for the full range of observations, and the range
of scale and type of data will determine the general shape of the chart
(see figure 6).

A lack of any rather uniform arrangement of the dots indicates
a lack of close relationship between the variables. '

A regreSSion line, or a line of best fit, may be drawn on a
correlation or scatter diagram. Values read from this line will give
an estimate of the change in one variable which is associated with a
change in-the other. This line of best fit may be either straight
or curved accordingzto whether or not the correlation is linear.

366
sigma slants.

Pictorial charts are composed of symbolic sketches of various
objects such as men, animals, houses, or sacks of grain, designed-to
give an impression of the magnitude of the items. These charts are
to be avoided, because there is eVery chance for error and confusion
in reading thenu Pictorial charts are generally drawn. in We and three
dimensions and the eye is not trained to interpret accuratelyitems deal-
ing in more than one dimension. , The very nature of the pictures will
give solid figures out of proportion to the data, and will serve only
to confuse and at times even to misinform the reader.

Mi scellaneous Charts

Numerous kinds of charts, made for various specific pur-
poses, are not in general use in research work.~ As a rule these
charts are a combination of the general types discussed.

Organization charts showing the organization of a unit from
the head to' the subordinates, calculating charts, three-variable cor-
relation diagrams, and many other types are used in specific cases.
Calculating charts and correlation diagrams must be accurate and pre-
cise. Their interpretation is limited to persons who are familiar
with the subject under consideration.

9.9.2 Mm

Dot maps are excellent for cenveying the correct impresaion
of "location" to a reader. is a rule, they are understood at a.
glance and are subject to few chances of misinterpretation (see
figure 7). ~

The effectiveness of a dot map depends to a great extent
upon the scale selected, the size of dot used, and the actual dotting
of the map.

In determining the scale to use for one dot, the range of
values in the data from the highest to the lowest must be considered
in order to have a reasonable number of dots distributed over the
map. If the scale is too large (that is, if each dot represents
1,000 acres instead of 100) the dots may be too few in number and
the map may not show distribution as it should. If the scale is
too small, too many dots may be required and variations in concentra-
tion may be lost because the whole map may be too black.

The actual size of dot to be'used is determined to a large
extent by the number required to be used by the data for agiven
scale and the space on the map available to place the dots. The dot
must be large enough to reproduce well, individually, and must be
small enougi so that they do not shade large parts of the map. too

,up--——

367

V-
heavily. It is necessary‘from the data to decide whether it will be
better to use a large scale and large dots or a small scale and small
dots. ‘ As a rule, small dots with a small scale are more satisfactory.
Large dots: and a large scale are used only when the dots are concen-
trated heavily in one area and are infrequent in others.

Summetrical even—sized dots are necessary to make the map
appear well balanced. Even contact of pen with paper with a rather
generous and even flow of ink from the pen will generally produce the
best results. A circular flat pointed pen is necessary. Dots should
not be placed in'rows or bunched to form a design of any kind.

. Attention should be given to the placing of dots in relation
to‘tOpography. In county data, where a. certain number of dots repre-
sents: some agricultural feature, the dots should be placed if possible,
in the areas to which the data applies. In mountainnus and rough
”3513, and in irrigated sections, the dots should coincide with the
actual producing area, if possible. Topographic relief maps and maps
showing physical, features are excellent guides in placing dots.

’ The number of dots on a map should be checked against the
original data. The use of base maps with blue county lines and names
is a great help in dotting the nap and in checking. The blue lines
disappear in photography and reproduction.

The scale should always be indicated on the map.
Hatched Maps

Line-shaded maps, showing sections that vary in density or
are of approximate equal density but of different design, are used ex—
tensively in locating data in its geographical position. The shading
of the areas is a distinct help to the eye in grasping the significance
of the data (see figure 8).

When a density shading map is made, with the density graduat~
ing from black to white, the most important areas (or the areas in which
the magnitude of the data is the greatest, etc.) are given the greatest
density. When a hatched map is made, showing areas of equal inportance
in a discussion, the density of the areas should appear about the same
to the eye, the design being different. A series of density and areal
shadings are given in Figure 10 as an example of the Various designs
that can be used.

Hatched maps of the United state's,de on a state unit basis,
are not the best for portrayal of agricultural statistics. The state
boundaries do not encompaSS agricultural areas in many cases. County
data, grouped to make larger areas and disregarding state lines, are
far more accurate in portraying agricultural conditions.

368

Hatched maps are difficult and expensive to make. A well-shaded
map, which will reproduce well, requires an eXperienced draftsman with good
equipment. A section-liner machine for drawing parallel lines an equal
distance apart is a great help.

Contrasts in density shadings are important. If too heavy
shadings are used the map will appear too dark to the eye and will not
reproduce well. All lines used in the map should be able to Withstand
easily the reduction necessary in printing. White spaces betweenblack
lines on a hatched map closed up appreciably in printing, and the repro—
duction of a hatched map as a rule appears darker than the original draw-
ing.

Original data for hatched maps should be subject to no changes.
Few alterations of shading can be made after a map is once dram.

A legend on the map should give the ranges of the hatchings
according to the data. As a rule the fewer densities shown, the more
effective the map. The range of data for each distinction should be as
consistent as poSsible with the range for the other distinctions.

No printing should appear in the hatching on a map except in
cases where the area is large enough to hold the lettering without de-
tracting from the visual effect.

0 ircle maps

The placing cf circles in the divisions of a large area is

used to portray certain features of an area... Circles can be placed
in the s ates in the United States or in counties in a state, he size

of the circle Varying with the magnitude of the data pertaining to the
state cr to the county (see figure 9). lthis type of map is not ac-
curate in interpretation because a circle shows an area which is two—
dimensional and varies as the square. of the diameter. The eye does -.
not make this comparison accurately. '

The size of circles to use depends upon the limits of the
space in which they are to be placed. If the circles are of sufficient
size, they may be segmented into component parts to show divisions of
the whole. Circles can be hatched or shaded to make individual pie
diagrams. -

As open circles are not striking to the eye, all circles
should be either shaded or black.

Miscellaneous Maps .,
Line and bar charts may be superimposed on a map to show

certain data in geographic position; movement or "flow" maps can
show the marketing channels followed by commodities; international

369

trade maps can show the movement of a commodity between Countries. As
a rule, these rather comlicated maps require skill in making if they
are to appear well in publiCation.

Identification maps and key maps are used to give the setting
or location of a particular area in relation to general physical features.
These maps are generally a combination cf :9. hatched. or dotted map and a
topographical map. The more simple they are made the greater will be
their value. ‘

Topographical maps, showing in detail the water areas, land
relief (mountains, valleys, etc.) in contour, and towns, cities, roads,
railroads, boundaries, etc., are valuable aids in all kinds of map making,
The U. S. Geological Survey at Washington, D. C., will supply on retuSt
detailed survey sheets for large portions of the United States showing
these tepographical features. '

Graphic Standardizat ion

Any organization that has uSe for various types of graphic
presentation should consider some general principles of standardization
to be used in the making of the graphs. If the original charts can be
standardized, the uses to which they can be put are greatly increased.
By standardization, here, is meant the operation of a system by which
the general layout, the sizes of all the charts and. naps, the sizes of
the lines and lettering, and the actual drawing of the chart in general,
conform to rather definite rules.

It will be recognized that uniformity and completeness in a
series of charts or maps, as finally published, add a great deal to the
value of the work. The uniformity of lettering and width of lines, and
the general‘eppropriateness of illustrations, add much to their effective-
ness ‘in a. publication.

All illustrations for publication should be made in accordance
with the reduction which will be necessary. Many charts are made which
will not be effective in reproduction; for instance, the lines and letter-
ing may be either too large or too, sml to appear well in reduction,

Z’iall charts or charts of large dimensions can be made under a.
standard system so that they can be photographed and reproduced as il—
lustrations for bulletins. It is logical that the larger the chart the
larger the lettering and lines should appear. If all the lettering and'"
lines on all charts, whether of Wall chart size or smaller, conform to a
set standard, the chart can be used for many purposes other than perhaps
at first intended. Small charts can be enlarged towall chart size and
vice versa, and. the result will always be a Well-balanced illustration.

The rec0gnized principles of-graphics are more easily adhered-
to if all charts are made under a standard system with certain set rules-

etc

A standardized set of instructions to a. draftsman helps to keep errors
at a. mi nimum.

Standardized graft ing Technique

To the draftsman who is actually making: the illustration,
standardization of technique is of great benefit. The draftsman, as
a rule, is not a. statistician and should be depended upon only to
make the chart as directed. If a carefully planned layout is given
him, together'nith all sizes of lines and lettering, he can devote
himself to the actual mechanics of the drafting. .If all charts, then,
are made by this same pattern, either by the same draftsmsn or by
several, the final charts will be uniform and will portray the data -
in the most effective way. .

tandardized technique increases the Speed with which the
chart can be made. The time required for a draftsman to determine the
general layout of the chart and the size of lines and lettering,r is con—
siderable. Nbreover, he often has'notthe knowledge of the subject
matter to determine the most effective style of chart to be made. AS
a rule, the draftsrran should not be required to know all the principles
of graphic presentation. His ,job is to follow directions and if a
complete set of directions is given him, his work is more efficient and
effective. A study of Figure ll will give one an idea how this is
done.

In the lower right—hand corner of this illustration is a.
reproduction of an experimental layout. Such a layout is furnished
to the draftsman with each job to be done. On this rough layout is
definitely indicated the sizes to use in making all lines and lettering.
The sizes of lines refer to the scale of lines furnished each draftsman.
The size of pen is then selected to make the right sized lines. Any
good set of pens having a full range of sizes can be used in the work.

If Figure l; is enlarged to three times the size here shown,
it will show the actual size of lines and lettering to use on charts
for one-half and full—page bulletin illustrations. The reduction lines
indicate the size the lettering will appear in print after reduction from
16-1/2 inch original to the 4-1/8 inches. The heavy diagonal line running
through the one-half and one-page layouts can be used toyindicate the size
cf lettering to use for any sized original between 16—1/2 inches and 4—1/8
inches. The size of lettering is shown by drawing a. vertical line from
the heavy diagonal line down to the dashed "reduction line", and measuring
the width of the Space between these two lines. ’

All sizes of lettering are indicated in the metric system which
is in use by the Coast and Geodetic Survey, which furnishes the standard
lettering sheets at a nominal cost.

In making mall charts or other charts that are larger than
those here illustrated, th standard sizes of lettering and lines are
enlarged in proportion to the size of chart made. In this way, no
matter what the size of the original drawing, it can be reduced to a
size to fit bulletin publication.

Maps and map charts Can be made to conform to this standard
by making the base maps to fit the requirement. Lettering, size of
dots, etc., are then indicated as on an ordinary chart layout.

All charts should be accurately checked when completed that
the portrayal may be exactly'in accorde cc with the data. Before
sending them to the printer, the size to which each chart is to be
reduced for publication is marked on the margin.

In an organization employing several draftsmen, a record
of the time required to mAke all charts will make it possible to
draw up budget of costs. This time record usually discloses the
fact that some draftsmen here a natural ability ta do certain jobs
Well, such as lettering, dotting maps, etc. Jobs can then be assigned
according to each draftsman's ability. In this way the production of
the unit is kept at a maximum and more satisfactory work is turned out
with fewer errors than when the jobs are assigned in a routine way.

Eraiiins.lssi2neni
In the drafting organization of the Bureau of Agricultural
Economics which does general map and chart work, it has been found
necessary to hare the following equipment for each draftsman:

Drawing table with cover and chair

set Bareh~Payzant pens, of Leroy or equiValent make
"T" square or straight edge

Triangles (45°, 60°, and 3 in 8)
Engineer scale rule

Lithographic needle (No. 4)

Protractor (100 divisions instead of 360°)
Bottle black waterproof indie ink
Penholders

Steel eraser .

Dozen thumb tacks

No. 290 Gillott pen points

No. 170 " " "

N O . 303 I! ll 1!

Scale of lines adapted to type of work
Samples of spacing for lettering

set of alphabets adapted to type of work
Set of calendar sheets, 1917 to date

Set of State maps with counties

Postal Guide

6H pencil

Hua Hideanikdnannm FIFJRDP'F‘H P’GIP’F‘H

372
2 No. 2 writing pencils
1 3H pencil
1 Pencil Pointer
1 Van Dyke eraser No. 6000
1 Van Dyke soft eraser No. 6500
SEmfllbbtmrs
l-Pen wiper
1 Pair 6” scissors
1 Roll adding machine paper

Special sized wall charts, display panels, and poster work may
require brushes and paints, but this type of illustration is not used to
any extent for reduction and publiCation.

Care should be exercised in the selection of paper upon which
to make original charts. This paper should have a smooth hard finish,
either cold pressed, calendered or fine textured, which will allow for
at least three erasings on the same spot Without wearing through.
Tracing cloth can be used for some purposes, but it is not as gener—
ally satisfactory as white paper.

Exagples 9: Types Q; Illustrations

The following illustrations Were selected as examples of the
more common types which are in rather generAl use. Specific reference
is made to the most important features of each illustration, with the
hope that these principles may be of use in constructing illustrations
of a similar nature.

Figure 1. Line Chart.

1. The lines representing data are the most important to
the eye.

2. The types of lines used are according to the relative
izportance of the data.

3. The base line is, next to the data lines, the HDSt ins
portant to the eye.

4. Grid line , forming the field upon which the data are
plotted, are limited to the most neceSSRry, are of even
weight, and are only heavy enough to reproduce.

5. The data curves are not labeled in the field of the
chart as this might interfere with visualization.

o. The weight of the lettering indicating the horizontal
and vertical sexles is only heavy enough to be easily
read.

Figure

Figure

Figure

7,

373 '

Scale captions clearly indicate the units used in plotting.

2 3116. 3. Bar Charts.

1.

1._

2.

3.

4.

4.

Time series bar charts generally are vertical except in
short series, whereas other bar charts are usually hori-
zontal.

The Space between the bars equals the width of a bar.

The zero base line is prominent. All other grid lines
are only heavy enough to reproduce.

The scales are clearly indicated.

Black or shaded bars are more striking than hollon bars.
There is no vertical lettering on the chart.

The scale unit intervals are so indicated. that the i
approximate lengths of bars in terms of the data can

be easily read.

Ratio Charts.

The general construction is similar to a line chart.

The 100 base line is heavy becauee index numbers are
plotted on this particular chart.

There is no zero base line.

The plotting of all data points whether falling on or
between the indicated scale’lines is on a logarithmic
basis and not on an arithmetic basis.

Pie Diagram.

Pie diagrams are not as effective in visualization as
one dimensional figures, such as bars.

The plotting of values begins at the top of the circle
and proceeds clock-vise.

The circumference line is heavier than any other one on
the diagram.

~ Hatched designs are used to define more clearly the

segments.

374

Figure

5.

‘1-

Figure

2.

3.

4.

5.

6.

Hatched designs varying in density from black to white
are used to designate the relative importance of the
data represented by the segments regardless of the .
size of the segments.

Hatched designs of approximate equal density (areal
designs) may be need for segments that are of equal
importance.

Different sized circles are made to represent different
magnitudes of data, the areas of the circles varying as
the squares of their diameters.

All lettering is so placed as to be most easily read
from one position.

Dot Map.

The allocation of dots is determdned by the data for
the smallest possible geographic unit for which dam
are available, regardless of the geographic divisions
shown in the final reproduction.

The dots are placed in definite relation to'topography
and other related factors,tdthin the geographic unit ”
designated by the data, such as township, county, or
state.

The scale was selected to give the truest visualization
of the data.

The dots are uniform in size and large enough to be
definite and clear in reproduction.

The dots are not placed in rows or to font any design.

The legend indicates What each dot represents.

Hatched Map. ‘

The density 5 hadings may from black to white in
accordance with the relative importance (or degree
of intensity, etc.) of the data.

' The black area covers a cemparatively small portion

of the map. The scale and shadings were selected
so as to bring this about.

hire

5.

6.

Hatched designs show clear distinctions.
The legend of shadings appears on the map.

The range of data for each hatched distinction is
uniform.

Shadings of approximate equal densities can be used
for data of equal importance.

The boundaries of the hatched designs are distinghishable
from boundary lines of geographic divisions.

Lines and dots in all designs are of sufficient weight
to reproduce and are far enough apart to prevent filling
in printing,

(Hatched deSigns which cause optiCal illusions are never
used.)

Circle Maps.

The areas of the circles vary according to the magnitude
of the data.

The range of sizes of the circles is dependent upon the
space in which the circles must fit.

Black or hatched circles are more strikin to the eye
than hollow circles.

There are no figures in the circles.

A scale of sizes of circles in terms of areas or
diameters is a help in interpretation.

Two—dimensional surfaces like circles are not accurately
evaluated by the eye, but circles are less misleading
than squares.

The circles may be segmented if large enough so that the
segments will show in reproduction.

.1mH"..uH;." 7 M
. n J '.'-‘ .3“

v ‘

‘13,; J1 ’

(art; 31"}. :5 *
ML“ .4:

 

376

COTTON PRODUCTION OF UNITED STATES. EGYPT. AND INDIA. I89I 'I926

SALES

 

MILLIONS

I

United States

 

 

'8

India

 

- - - Em”

 

/

 

I 6

 

 

 

I 4

 

 

 

 

 

 

 

 

 

 

 

 

A A A |

 

 

 

 

 

A l A L

 

A A A A A

 

 

l I A A A

 

l890

I9I0
FIGURE I

I895 I9|5

I920

I925 |930

VOLUME OF BUSINESS AND COST OF HANDLING GRAIN. I9| I-I2—I925-26

THOUSAND

CENTS

 

BUSHELS
HANDLED

I80

SASKATCHEWAN COOPERATIVE ELEVATOR COMPANY. LTD.

 

 

I50

I20

90

 

60

30

 

Volume of business

I9ll-I2

A

PER
BUSHEL

 

 

\‘_ _.__

 

Hand/Mg cosf \
\ \

 

ISIS-I6

I920-2l
FIGURE

\
II \
I \
I \
TI
I \
I \

__\

6

 

l925-26

 

BELGIUM --------
UNITED STATES--
UNITED KINGDOM

RUSSIA(EUROPEAN)

GREECE

ARGENTINA ------
BULGARIA '''''
INDIA '"

RUSSIA (ASIATIC)"

INDEX
NUMBERS

200

IOO
80

60
50

40
30

20

I o
l865-‘66 I875-‘76 1885-86 I895-‘96 I905-'06 I9I5-Is l925-‘26

3'77

CULTIVATED AREA PER WORK ANIMAL IN SPECIFIED COUNTRIES

NUMBER OF ACRES
Q 5 l0 IS 20
l |

 

 

 

25

 

 

FIGURE 3

TREND OF PRINCIPAL AGRICULTURAL EXPORTS. I865 ' |928

YEARS ENDING JUNE 30
- = ALL COMMODITIES
l9|0 |9|4 IOO c/ Conan

ALL COMMOD/ [5

"—I‘ I’D ’ ‘_l
- I 'I
’ ‘\- ’ 4113/
I I
’4’,»

COTTON

 

I870-‘7l I880-‘8l I890-‘9I |900~'O| |9|0-'|| I920-‘2I
Fl GU R E It

I930-‘3I

afimyudxz 83%»;

,:.,: “.1“; :....,,,. «”1“... ‘ ' , 1 7 .. .4 I _
' f Mr“ ._ ~ -‘ .1- _ . ; ,. [rah

F

maxim}-

m,“ ; _.'._.-.._ .‘

oer-3:1.“
. 3 {f1 PM?“ ‘ mm '1
. s:\-{‘.‘fi¢‘ #17,”;3 u “‘“"'" ’, ,. ,

 

 

 

 

 

 

 

 

. M Ww 44»:
:8v1"?-~.;:« ~

~

 

85.2

 

 

MIXED WHEAT

ALL DISTRICTS EXCEPT THOSE OF PACIFIC
COAST DIVISION

WHITE CLUB 0.1%

  

 

 

 

 

 

 

 

 

w.c.TRACEx
c.w. 6.31 , lI-I.Ia.s.7.3%
.a.‘ ' I!
’30.! I ':
[/llll Ru“
.7:
.n. .
32.0
RECEIPTS 82.67: SHIPMENTS I141

PACIFIC COAST DIVISION

. .

‘- 'ch'MbN' wn'ITE_'.'-,'.
- ' '_ 457.87..

 

 

RECEIPTS 88.87, SHIPMENTS ”.21

 

FIGURE 5

 

~ "_ . ‘j' a. A .
SEWER»: .4!“ .L‘sé-i :17 :7“

. y“.

.i. -- ,"

‘ ‘ . 4 4!.
aim-1x1!” tan-guy”; [“9“

 

1.. (

 

 

 

5’79

RELATION BETWEEN COST PER POUND OF GAIN AND PROFIT PER HEAD OF STEERS

PROFIT
PER HEAD
DOLLARS

   
 

48

32

. ' ' nodrrr nu pun-Ma,“

ofcasr per pound

mmmtxuum . ’ ."

r. I -.6558
b - L78 a.

. Una
'28 Promo/.75 Coarv

llo I8 22 26 30 34 38 42 1.6 so '54 58 62 66 7o 74 78 82 86 90
COST PER FOUND IN CENTS

FIGURE 6

 

    

VALUE OF CROPS. 1921+

 
 
 
 

Omng ra change: in the crops Inc/udod
and 1/1 price lava/s, flu: map it no! :fricr/y
comparably WI"? Marci Value afA/l Crap:
in (9/9 Each dot rep/mun"

1% 1‘ one million dollar:

    
 

 

 

 

FIGURE 7

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

, . v, .». 1,1,“.
MHz—”(Quwvfi‘ ”A“, a; -

 

 

 

 

 

380

 

      

FARMERS HARVESTING WHEAT
PERCENTAGE OF ALL FARMERS. |9|9

/2o-39

   

 

 

 

 

m“ - so
“so -69
-800ndavor
FIGURE 8
J33 SALES AND PURCHASES THROUGH FARMERS'

BUSINESS ORGANIZATIONS, I924

    

 

AREA or 9 Rths LLLLLLLL

AI
VA LUE EQUIVALENT OI LLLLLLL

 

 

 

FIGURE 3

'.,~

v V

2"“ 5a.... 1.. .. a»
v ‘ ‘ »' " ‘

A

.v 1

- \
"'5’.

- M:
i" ‘/

 

381

 

AREA DISTINCTIONS

 

 

 

 

 

 

g;
Ix
vv‘, ”‘4‘
< x76

v
x

v

. .
"vxaxw

v

111111111

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

DENSITY DISTINCTIONS

 

 

 

 

 

11111111

11111111
11111111
1111111/
11111111
11111111
11111111
11111111
11111111
11111111

 

  

 

 

  

 

 

 

 

 

 

 

     

.1-1.1-1-1

.1-
,.1

  

v1.1.1.1
I 1.1.4
.1.1.1.1

 

 

 

 

 

 

FIGURE IO

3“

An:

, (.5.

 

 

 

 

w- ,a,..‘._ ,

' . 1 M

‘ ’ .. , ' K

,. ‘ u * ,x,

' "“ ””2 Nb" ‘0”
”"' ’ “31%

'_l:'d"»a?"». ,K: 1‘ _.

x

 

who

A

 

 

 

 

 

 

 

 

 

         

 

 

: HIDE—u
dukfifhbqi MIN! h< isaflk‘g 025; 1.. l..un« .
1.33am ukhxn Qhkk<3 .hutikot‘ufik \\\§<z§§=§<ti‘~x¥.30m<‘ .. ‘ ‘ I H .
35‘ . uncouth .uskukaq .3 ~30 At! .125 \ . .
.45 #33. SM 139.: «Issuilulku at» . x I ‘‘‘‘‘‘‘‘‘‘‘‘‘‘ un‘l
. __ x ......................... .. o m < . . .............. _ . n \
\ ........................ 2 U m < x ........................ {El 3&5“!ng «SSERES
\ _ .wmtbQN N\¢ Bx \umbfiafi
‘ I'Ijuh!
\ ....................... on U m d. ....................... l.
\ .
‘
‘ \ ........................ on U m (. .................................. .
‘ u x.
\ .......................... 9 on: .................................... ‘ n x "r. .RQ t—
X ....................... a ui.‘ ................................. u H \ .
Knit-1“! I: T. 3 3.4. ‘ ........................... . x
. . x
_ usual:- 3 \ w / u \
. .
. h 55.: to Mt: ht: but: 553)?! u s
. . 8k 33}: m4 Kass; . x
u N 95393 a: g .{wfi H x
. Manila... .3: .G 9333:?
FDO><J J<FZUI.KWn-XM MIL. m “ M433. 015$an .45 \\
b \
its Ag .32 SB! Rafi-O;
3.1: win .8 no two. to tu3h~§ \
{vim xn HQ“ hat: meal to OIL \
.2: Eggs \3 .3 SK usuukkw \
~93. anti .3 M399 . 13:3 .3 g. N
\ x x \ \ ‘ x X \
\ \ I \ \ \ k \ \\\ \
\ \ x \ I x
r \r s v/ .25: tote—Vow.
\
fl , , \
W , V ~
, . _ ~
~
\
N
\
. \
\
H. \
s

auumm can nz< .mwzj ...:0.u.$

510523 KOk aw: w¢< MZM

  

 

 

m 02.1wtm4 L2<N><m. TI wthSfiE 55x3 mu: K 4559.6 1%
FDOxrj J<PZU<§KNQXU mIk

 

OIWW§U>§ gex .S§x‘4 NQ‘k \ d<§\Q\QQ 44.

OPP/«.5344. zofigomm .._o m._n:oz_xa I mmz... oz<.mzma.mSo><._ am<oz<hm
MDgZIOm—F EEOEZD < OZEDOm—m m0". DOIFw—z

 

 

 

, v . mm." A
.H .. , H aufiawxwr‘.
rag!“

V} .

u. .. . . . V 93?».

a. s A

«VI

. h.

«I
r u

 

kgflm»
. ,

.3; A“

w \.

“a; .

1..“ .4.) ~<fi

 

‘ ~ (2.1V

.6815

.6. Preparing the written Report“

The plan for the report of an investigation or a study begins
with the planning of the study itself, for the work of the modem
scientist is not done until the results have been made of use to his
fellow men. This preliminary planning affects the work throughout,
for any marked deficiency or defect in the original plan, unless re—
ctified early in the study, is inevitably revealed. in the final report
or bulletin.

In the course of the study much material is gathered and many.
obsarVations are made that are not called for in the plan, and the plan
itself may be improved and reshaped to a certain extent as the work
progreSSes. _It is probable that the investigator can scarcely learn
too much about his subject; yet a knowledge of there to stop is almost
as necessary. as a knowledge of where to begin, and this is as. true in
the preparation of the report as in the conduct of the study.

Discrimination a Basic Need

Discrimination in these matters is difficult for those en-
thusiastic workers who grow so absorbed in their investigations that
they are likely to lose perspective and to feel that every supplementary
detail is important. Thorough shifting and evaluation of results in
relation to their value to the public are prerequisities to a Success-
ful start on the actual preparation of the publication. Much that has
been useful to the investigator in his understanding of his subject and
in his interpretation of results is not essential to the report, and
might better be placed in office files for the use of any students who
wish to go into these details.

Orggpi zat ion Clarifies Re sults.

With this evaluation, comes the drafting of the definite
outline of the bulletin. Outlines must take into consideration the
space at the disposal of the author, adequacy of treatment, logical
development of subject, and proportion. Outlines are guides and
assistants, not masters, They may be altered and developed as the
writing proceeds, but a well—made outline following a carefully planned
and conducted study is much the same before the work of writing‘is begun
and. after it is completed.

Investigators sometimes find that when they have completed a.
study the results at first seem to form a rather incoherent mass in
the files and in the mind. The selection, systematizing, and clarifica-
tion should come before the actual writing.

 

* By Caroline B. Sherman, Bureau of Agricultural Economics.

-v.‘

384
Character of Publication Settles Many Points

Arrangement of the outline and material differs with the kind
of publication. For the scientific journal or bulletin the accepted
order is followed: (1) Review of literature of the subject, (2) con—
ditions under which study was made, scope of study, and methods used,
(3) results ofstudy and their significance, implications, and relation—
ships to other aspects of life, (4) summary and conclusions, and (5)
list of literature cited, or a bibliography.

, In popular bulletins and articles on the other hand, the news
influence is now evident. 2‘. "lead" is used uhich will seize and hold
attention and will awaken definite interest in the story to come. Sig-
nificance of results and relationships are likely to find a place in
the opening, paragraphs of the popular scientific paper, followed by the
story of the Work. The review of others" work and the basic data, if
used at all, find a place well toward the end of the story. The casual
reader'of a popular article must get the gist of. the paper in the opening
paragraphs; the scientificreader of a scientific paper expects to get
his story'logically and, in a naturally progressive way. But there are
increasing sighs‘that real scientists are recognizing the possible value
of the opening paragraphs in arresting attentionandawakening keen
interest.

Scientific Text Demands Accesgpries

Tables, charts, footnotes, and citations are reduced to the
minimum-in all popular or semi-technical writing, but are vitally
important in the scientific paper. But even here it is easy to over-
use them. Again discrimination is needed, and perhaps some degree of
relentlesaness, for scientific and economic authors seem particularly
loath today to limit their number of graphs to those which definitely
clarify or erphasize the important parts of the text.

Good tables have clear, concise, but adequate legends, have
specific and accurate stubs and column headings, and they read from
left to right. They are planned and constructed with'the constant use
of the author's critical faculty. They clarify the .text and are used
to secure concise reSults. Therefore good tables are rarely complicated
or difficult to follow. The salient facts in the table are stated in
the text, but the reader must go to the table for details. is a mole,
the tables represent all ofthe supporting data that the thoughtful
reader needs in order to be convinced by the text. " ' ‘

Good charts are simple, clear, and not difficult to follow.
The earlier idea that the more that could be shown-in the Space, the
finer the chart, is fast disappearing. A clear but adequate legend
(as short as practicable) and an adequate key are essential- The addi—
tion of an interpreting lower legend is appreciated by most readers

'385

even though the salient points brought out by a chart are repeated in
the text. Photographs are usually considered preferable to charts for
use in illustrating popular paperS.

Even scientists do not seem to like to read footnotes; there—
fore their use is now more limited than is true in the case of charts.
If the material or statement is merely parenthetical the text can often
carry the statement that previously would have gone into a footnote. If
bibliographic references are infrequent they are used as fbotnotes; if
they are extensive they are grouped in a list at the end with symbols
for reference from the text.

A bibliography is usually more extensive than is the list of
the publications to which reference is actually made in the text. It
usually aims to be a well—rounded and reasonably complete selection of
recommended readings on the subject. Forms of citations have been
rather well standardized during recent years. Many authors now follow
the forms in arrangment and abbreviations that are used in some gener-
ally read and accepted journal such as the Journal of Agricultural
Research or one of the journals of abstracts. In any case citations
should be complete, concise, and accurate. Accuracy necessitates a
verification shertly before the bulletin goes to press, for reworking
and copying are likely to adnit errOrs even if the original jotting
down of the reference was accurately done.

Summaries and Conclusions Important

Summaries or conclusions, or both, are expected by all
readers. In popular writing they may be implied or lightly sketched
in, but in scientific :miting the summary contains an adequate digest
of all of the most important points covered by the bulletin and it
contains nothing not found in the bulletin. The conclusions, on the
other hand, may show deductions drawn by the author to which the reader
is not obligated to subscribe. From the text, the supporting data,
and the summary, the reader may draw his own conclusions which may not
necessarily coincide with those of the author; but if they are clearly
differentiated from the summary and are conservatively and thoughtfully
drawn, the author is usually entitled to formulate and present his own
conclusions.

Clarity and Brevity Prerequisities of Successful Style

Style in writing is determined to a considerable extent by
the nature of the bulletin and the kind of readers to be reached. The
more scientific the subject and the probable readers, the more dignified
are the style and diction. Popular bulletins and articles on scientific
subjects are more conversational, both in style and diction. In either
case the successful style is clear and concise.

Statements must be so made that they cannot be misunderstood.
It is not enough merely to have them capable of understanding by the

it's/s j

careful reader. This means the exercise of the author's critical .
faculty throughout the Writing of the paper. This clarity is obtained
largely through clear thinking, brevity in statement, and exact use of
well-chosen words. If a writer does not have a thought clear in his
mind he will not make it clear on paper. Even when the thought is
ones clearly stated, it is well to cut out every superflous word and
phrase. In scientific and semi—technical writing cutting usually
means a. pointing—up and an erphcsis.

Clarity and brevi+ y make for short papers and bulletins,
which make for more readers. By each page that a paper is shortened
readers are gained.01arity and brevity mean short, well—knit sentences
devoted to single thoughts and ideas, and. building a'short, well—knit
paragraph. Variety 1n structuie ofs entences can be secured without
losing the effect of brevi «y. Clarity is aided oy frequent subheadsa
by precision in use of terms, by grief encplanation or definition of
unusual terms or terms used in an unusual way, by the use of concrete
rather than abstract words‘and terms, and by the use of occasional we11~
selected exanples or illustrations.

Further Lids are Needed

Clarity and brevity do much to make a text interesting, but
further aids are needed. Us inf; Significant places, as at the beginning
and end of paragraphs and of sentences is one aid. Using the active
voice and present tense whenever possible is another. The tempo, or
rate of movement of the author's style,‘has much to do with holding
the reader‘s interest. ‘11 slow-moving, heavy- footed text may carry the
patient reader through to valuable conclusions, but an alert, quick
style, if not overdone, will cairy both the patient and the impatient
on to conclusions 7hich can be .just as sound and valuable as those which
have been reached by plodding processes. -

Final Responsibilities Differ in Different Cases

Conditions in different institutions differ widely as to the
amount of responsibility the author must asaume for final editing.
checking of tables and graphs, verification of citations, and reading
of proof. ' If such work is not provided for by special workers the
author‘s responsibility is great, for upon the accuracy of these details
often depends the final Value of his work. Untrue swmnaries and un-
sound conclusions can be drawn from inaccurate data. Even in those
instances inymich special workers are charged with these details for
all publications, authors usually find it instructive and profitable
to follow‘the work through to the end. -Some leaders hold that no in-
vestigator's work is finished until he has thus followed it through.
Certainly many an institution must have been surprised, if not dis-
mayed, by the manuscripts submitted by a new worker, if he has been
chosen partly for his written results as published by an institution ,
that maintains a strong editorial force.

.1

-' 2375?.

Whether such aids in editing and checking are available or not,
the author owes it to himself to allow a short lapse of time, and a certain
preoccupation with other matters, to intervene between his completion of
a manuscript and his final critical examination of it. The final review
may well take the form of a quick cross—examination of himself as to selec—
tion and elimination of material; arrangement and statement of facts; use
and form of tables, illustrations and citations; choice of words; termin—
ology employed; and conclusions drawn. The author who has thus met close
criticism of his manuscript is likely to give a vaIUable report to the
public.

D. Utilizing the Product of Research*

In this section an effort is made to report the experience of
various agencies in utilizing the product of research in agricultural
economics. Much of it is a mere description of rather common practices,
but included with it are reports of experience with Various practices and
probably a few suggestions of practice new to some. No general question-
naire was sent out on the subject, but nearly a score of people who have
had close contact with the problem contributed reports of experience and
ideas which are here included without Special reference in most cases.

The outline followed is according to uses rather than to media
of publication, but under each head the appropriate media of publication
are analyzed according to their suitability for the use in question. The
various uses considered are as follows:

1. In the class room
2. With farmers and small marketing units
3. With large business organizations
4. With public agencies
5. In connection with other research
The various media considered are as follows:

1. Class room discussion and accompanying reading
exercises and problems

2. Textbooks and treatises
/’
3. Bulletins and circulars of various types

4. Scientific articles

5. Articles in the press

 

*Mr. Tolley and Mr. Tapp assembled much of the material of this section.

388
6. Short courses

7. Field courses

8. Posters

9. Exhibits
10.. Demonstrations
ll. "Drives”

12. CorreSpondence courses

13. Service Work — installation of accounting systems,
accounting analysis of records and the like for
individual farmers and other agencies

14. Personal work with individual farmers and the like
15. Talks to groups

16. Slides

1?. Addresses at public gatherings

18. Conferences

19. Periodical letters, etc.

20. Working through farmer organizations

Utilization with farmers A large portion of the research

or small marketing units work in agricultural economics has been

initiated on the assumption that the re—

sults would be of direct value to farmers as a basis for improving their
situation, either as to the.production or the marketing of their products.
The adequacy or inadequacy of the various means whereby the results are
made available to farmers is therefore a matter of considerable interest
to research workers.

The most general channel through which the results may be ex-
pected to reach farmers is that provided by the agricultural extension
service. Experiences in this respect have been varied. A few people
reported the failure of the extension service to use the results of
economic research as the greatest handicap in obtaining the utilization
of these results by farmers. Others commented upon the necessity of
a close working relationship between those doing research and those
doing extension work. Some raised the question as to the extent to
which workers who are not specifically trained in agricultural economics
can be relied upon to aid in conveying to farmers the results of certain

types of economic research. Others would have the entire extension
personnel develop an economic approach to their work which would be
based upon and utilize fully the available results of he economic
research which were applicable to the problems at hand. The latter
plan in so far as it is possible provides a broader channel through
which the results may be disseminated to farmers. In certain fields,
however, the research workers must be closely identified with the
program of getting the results to farmers. This is particularly
true in the field of price analysis where the results of the research
to be useful to farmers must be interpreted currently in terms of
current and prOSpective conditions. If the members of the extension
staff generally are to participate in this work, there must be the

‘ closest kind of cooperation between the economic groups and the other
groups.

In a few cases the results of economic research appear to
constitute the major basis for the extension program in its broader
aSpects. In other cases where the research has not been done, an
attempt is being made to substitute a hurried economic analyses of
the situation for the broader background provided by research.
Available infonnation on botl types of arrangements suggests that
the tendency in the making of extension program will be to place
greater emphasis upon aiding farmers in the use of the results of
economic research and of current economic information. Such a devel—
opment would obviously place research work in agricultural economics
in a more strategic position from the standpoint of the use of the
results by farmers.

Media through which farmers are being reached with the re—
sults of economic research, either directly, or indirectly through
the extension agents, are discussed below.

(a) Bulletins

Two ideas seem to pervade present programs of bulletin
publication for farmers. One is to publish short bulletins covering
one or a few specific points; and the other is to publish long bulletins'
giving useful infopmation on all important aspects of the subject, it
being assumed that these will serve as handbooks for ready reference.
No doubt both of these sort of publications have their places, but the
first nf these practices is by far the predominating one. Farmers’
bulletins are becoming more clearly differentiated in type, are becoming
shorter, simpler, more readable, with more illustrations, more attractive
covers. They are being prepared more with a view to including what the
farmers need to know rather than what happens to have come out of a par-
ticular study. Oftentimes they present the results of no Special study
recently made. Others of them are pepular presentations of parts of
technical bulletins. ’

890

The extension circular or leaflet is still further removed
from current research, although it sometimes grove out of it. Its
essential features are timeliness, brevity, and wide distribution.

If we will add the further circumstance that technical
bulletins come in degrees also according to how difficult the sub-
Ject is and how fundamentally it is studied, we have the array of the
publications in pamphlet fonn. The most severe criticism that can
be made of them is the confusion of types which many of them represent.
Particularly a large proportion‘of them are too difficult for ordinary
farmers, or contain a great deal of data that farmers will never make
use of; yet too elementary to represent any significant contribution
as a research bulletin. Many of these had better be pruned severely
and published either as one or the other. Some technical bulletins
are properly too difficult for distribution except among co—workers.

Bulletins for other groups, such as elevator managers, cream-
ery managers, bankers, etc., can be more technical than'thOse to fanners
in general, but not greatly so. Such men are likely to be more highly
Specialized than farmers, and thus capable of handling more technical
aspects of their business, but only to a degree.

Many of the correspondents expressed doubt as to how much
farmers' bulletins are read, and some even wondered as to whether they
are worth publishing. Yet the facts are that the number of inquiries
that come in on many subjects is sufficient to warrant preparing printed
Statements as answers. This much done, it seems only economical to
give the publiCation a wide circulation. Several expressed the judge—
ment that there were enough college—trained men in the states alone to
warrant the publiCation of technical bulletins for their use._ No doubt
some economies could be introduced by more careful selection of bulletins
and supervision of mailing lists.

In various sections following will be presented several arrange— V
ments for making bulletins more effective.

(73) 1mg

Azticles First of these is the use of the press ;_dailies,
rural weeklies,-and farm Journals. Several departments'reported pre—
paring a series of short articles just at the time the bulletin is to
appear. This.brings in a large volume of requests for it. Others
supply the press with these articles in adVance of publication as
bulletins, realizing that the press greatly prefers this arrangement.
Another arrangement is to distribute abstracts of bulletins widely in
advance; another to send out brief press notices of them. The ad—
Vantage Of these articles is that they are likely to be more Widely
read than the bulletins. At least, this Was the testimony of these

391 4,

reporting-on the subject. Then if there are some who want more details,
they can always write for a copy of the bulletin. Both these and the
publication or press notices have the effect of interesting the people

in the bulletin before it appears, so that they are likely to read it
when it 'comesy' One publication department stated that no bulletin

was ever sent out until a campaign of publicity had sold the bulletin

to the farmers in advance. The device of sending out notices of
bulletins or abstracts first to see who wants them has much to recom—
mend its

At some stations the publications department does a more
satisfactory job writing the short articles for the press; at others,
the economics department. Cooperation is desirable in any case,
capacially if the economic department has no staff members with Journal-
istic ability. "Loose talking" in press writing Was cautioned against
by several, . n ,

(0) Serial Letters,

. etc. Four correspondents reported that their depart-~ -
ments were issuing monthly letters or summaribs of economic information
of onekind or another, Cornell has demonstrated how successful this - "
can‘be: ‘ over 3 ,000 farmers in New York receive it.

(d) Lesson , ‘ ' ‘ - ' "
Leaflets In‘a' few cases the scheme has been followed of
preparing a dozen or more lesson leaflets and sending them out one at
a time atintervals ofia week or so. The judgment is that this is :
effective, but entails great expense'of labor andmailing. 18 part of
an important drive, the'expense‘ might'well be warranted.” . ' ‘

(e) Boaters The most. successful efforts along this line were ‘ .
posters containing information so useful that the farmer would nail
them in a conspicuous place - usually in his barn or his granary; or
that an Igovator manager or buttemaker might similarly use.

(f) flibitg In the first place, economic material is harder
than most to, display in‘ such a way that it will catch and holdzth'e‘
attention of passersby’ at fairs and exhibitions; in the’ss‘cond place,
serious question is raised by many as to whether any but the most"
spectacular material is worth presenting at popular exhibitions.__

1,..,

(g) Talks to " .1 _ " . . '
Small Greys One' of the conventional extension methods
is surely that of talking the material to small groups of farmers in
school houses, town halls, etc. No doubt it has been nearly as much
used with economics material as with any other kind. Few of the
correspondents had much to say in is favor. They did say that 1’0‘
often does reach the leading spirits among the farmers, and that die-
c'ussions started at these meetings take a firmer hold than those
started from reading because several get interested in it at the

392

some time. , More of the correspondents were concerned with ways of
modifying the__r_aeetin_gs so as to make them more effective. One of the
commonest procedures is to. arrange the meetings as part of a campaigl
with a specific, objective and in a given area. All, the devices for
publicity may be concentrated on this ccfipaign, so that the meetings
are thoroughly advertised. Additional features more spectacular in
nature are: sometimes added to bring the crewds. In 'one state, a
widespreadsurvey of the problem precedes the educational campaign.
Every producer of the product in the area receives a"copy of the
report of the study before the meetings are arranged. Conferences '
with leaders in the area are held when the survey is. planned. More
usually, the campaigns are less vigorous than this, being confined to
a series of meetings scattered ever the area. Sometimes a sequence
of several are‘arranged at each point. Charts are used freely in
present‘ingflthe material. .; ' ’ I. ' " ‘

(h) Addresses Addresses before larger gatherings, as at con‘-—’
ventions, were deemed Valuable as to opportunities for acquainting .
farm. and. community leaders with, the results of research and-preparing
the ground for later Work. “Several stressed the importancd of having
such addressee carefully prepared and :it was pointed out that"'the '
economist had better not talk at all then to talk lo'osely "or inSinoerely."
Apparently there is danger that those who can make the best addresses

are also capable of talking most loosely; but, as one department head
empressed it "The birds always come home to roost.“

(i) Individgg; ‘ ' . _ _

g , 1 . 17911:: ' ' Dr. Young of Vermont eXpreBsedthe prevailing ,
judgment on individual work: "Personal contacts are 'g‘ood',j'_but much
like trying to dig'the Panama Canal with a spade. The total accbmp-
lishmentis disappointingly meager and slow. Much ofthis method of
spreadingresearch results is a great waste of time." T Others pointed
out, howeve , thatwork with caremlly‘chosen persons, those'in a
position to carry, new ideas to many others, oftentimes proved tbgé
best possible way to get certain kinds of thinking started. County
agents, officersof cooperatives and the like,«are always worthy of
individual :work. Work with individuals is of course moh._moro
promisingof resultswhen there are fewer units, as in the case of
marketing. ' V '

(3) Service'

Work: [the line between service and extension pork .
is hard to, 5-1792“): and probably. it is.,not necesaary to draw it closely.“
Installing.) accounting systenfi in Creameries and keeping them going for
a year or two may be‘service work of the purest type, but it may be . ’1
justified as about the only successful Way of getting better‘ accounting
adopted- It is so costly, however,.that nest stations‘have had to
abandon: it. xfiith'farmers, the' situation is different, since records
will do for-mostpurposes, and these are easily introduced and supervised.

. ,_.,.

395

Service work to large cooperatives in the way of installing; systems of
reocrds for statistical analysis, or analyzing some of these data as a.
demonstration, is amply justified on the score of the number of farmers

indirectly reached, and for a few years to {get the work started.

(k) Qemonstrotiogs: ”Forking closely with a small number of farmers,
or even a single cooperative, ii. order to demnstrete the effectiveness
of certain methods or changes is often feasible and warranted in spite
of the expense inVOlved. In some cases the demonstrations are carried
on in connection with the research or as incidental to it. In other
cases the demonstrations are put on by the extension service but based
almost wholly upon the results of a research project ~ in one state,
for example, recommendations growing out of a research project served
as the basis for Specific changes on 20 demonstration farms in the

area studied.

(1) W: The importance assigned to short courses'en the
cmnpus by the correspondents is in the access it affords to leading
farr..ors, and the publicity it gives to the research work. Farmers‘
week is a. place where "thing/3s can be started." It also gives the
stations contact with the various fem oz'gmrizatim‘s that meet at the
same time. Rather indifferent success was reported with attempts to
{jet groups of marketing men to the campus for accounting: Courses and
the like.

(m) Field (Eggs: Apparently many insti‘mtions are trying the plan
of signing up farmers, or the young people on farms, for regular courses
711ich meet one a wool: and at which a definite course of instruction is
diiV01‘1. The reports as to its success are cenflicting. No doubt the
maxi-Liner of handling it melces Host of the difference in results. It has
much to recommend it on many counts if it can only be made to work. It
should be tried out thoroughly as a plan for teaching; accounting to
managers of smell marketing units.

(n) Correspondence Course; 'Ihus far, correspondence courses have
not made much headway with farmers, but considerable in the commercial
field. Agricultural economists should try them out thoroughly in
connection with accounting; and a few related lines. As a way of
reaching: form boys for other sorts of economics instruction, it should
be given a thorough trial before being abandoned.

(0) Conferences: The ”econordc conference" has recently come to
the fore as an instrunent of progress in getting the results of economic
and other research prominently before farmers and their leaders. The
essential idea of the modern conference is to call together a group of
the dblest men interested, present to them carefully prepared analyses
of all the data available and let them talk themselves throueh'te an
agreement as to a plan of action.f "Within the last five years es-
pecially, conferences of this kind have been held in many states, or

394

in parts of states, at Which state or regional pregrams or outlook
statements haye been prepared. All of these so—called economic con—
ferences have not, however, utilized in a large measure the results
of economic research. In fact, some have probably been more useful
in indicating the need for more research.

Some of these conferences have afforded an excellent means
of getting.the results of economic research before leading fanners in
a convincing manner. They have furthermore afforded the institutions
sponsoring them an opportunity to synthesize the results of econonfic
research with that of other fields in Such a Way as present to farmers
a balanced extension program, the lack of which is often confusing to
thoughtful farm leaders. Adequate preparation for such conferences
requires the closest cooperation between those doing economic research
and those doing research in other fields as well as with the extension
specialists. The experiences of several states suggest that economic
conferences of one kind or another may play an increasingly important
role in connection with the utilizations of the results of economic
research. '

Another type of conference is that at which a research project
is planned and organized, or the results reviewed before publication,
partly as a device for securing publicity and support for it. Another
advantage of it is that it brings together on each question the repre—
sentatives of all the departments of the institutions and all institu-
tions in the state.

(p) working with farm organizations: hention must be made of the
success obtained in many states with working through farm organizations
0f various sorts — federations of farmers cooperative organizations,
and general farm organizations. Studies have been planned with them
in advance, and schools and district meetings arranged subsequently.

One device which has been used with local c00perative associa-
tions to a limited extent consists of providing some method of accredit—
ing those units which maintain standards that research has shown to be
essential to success. This serves to center the attention of the
members and managers upon the standards set up as the result of research
and in that way aid in the improvement of the organizations.

(q) UtiliZation of research product

with large organizations: This section has reference
particularly to working out research programs with the large business
organizations in the state, cooperative or otherwise, with whom the
farmers have business relations, and subsequently getting them to profit
by the results. The reports on this subject were meager. Very appreci-
able headway has been made in scme states; little indeed in others. Too
frequently the principal concept of research that the officers of such
organizations have is the digging up of Some data, preferably true, that
will aid them in confounding their enemies. More frequently than not,

'r’

395
their recOrds "are not complet e'enou@ to allow statistical analysis of
their operations., The first step is to analyze what records they have
and show what more can be done with better records. While waiting for
new records to accumulate, it may be possible to arrange field. studies
that will aesemble other useful data. Some departments that are
working closely with large organizations in their states may be doing
little more than enjoying the privilege of sitting in with them in
council - Worth while but not contributing much to the progress of
the organisations.

(r) Utilisation with public agencies: In general, public bodies in
the United States make. less use, and feel the need less, of assistance
from reseaICh workers, than in almost any Country. The reasons for
this need not concern us here. The question is: How can this situation
be changed? The judgment of those who have made efforts along this
line is that the majority of public officials welcome data and analysis
that can be put to use and that in particular, legislators welcome
material that they can use in debates "and in analyzing public issues.
Probably if more agricultural economist weuld'ta’ke time to analyZe
public questions carefully, and publish their analyses and make them
aVailable to public men, they would be surprised to see how freely
they would be used. '

But ordinary discussion of such questions, such as economist
commonly indulge in the classroom, or among themselves, will not help.
Whatever is prepared for such use must show the character of genuine
research; and research on such problems is particularly difficult, and
not many agricultural economists are experienced with it.

Problems on which public agencies would be most likely to
welcome assistance at present are those relating to taxation, credit,
transportation and possibly the tariff in relation “to agriculture.

Use Of Un— ‘ The problem of utilization of material _ p '
ublished ‘ _ 1 . while it is waiting publication often 66- ~ '- ‘
Materials“ [15‘3" comes of great importance. one department

*- " ' - head writes: "With referenjfie to the utilizsi; .;
tion of unpublished materials and results, the research workers in some
cases get out to address farmers meetings, though this is rather in-
frequent. 'me extension division plans to utilize through its farm
advisers any information which may be available, and each fall holds
an annual conference at' which the research staff of any, division is
given an opportunity of presenting any material when such h‘as been
accumulated but not made available through publication." Another
department head is even more Specific as to the importance of un—
published research material as working stuff for the extension man.
‘w background has largely been in the extension field. As you 1010.";

 

’*This topic discussed by Professor C. L. Holmes.

396

I Was in extension work for more than ten years. Consequently I see
many things more from the standpoint of an 'extension program rather
than the idea of- 's'cién‘tific papers which add to the sum total of human
knowledge of economic affairs. I therefore believe in releasing
material for extension use while the iron is hot. I could cite several
instances of perfectly legitimate and practical projects which have been
well executed, but time between completion of the project and publishing
of the manuscript Was so great that the bulletin ceased to be what the ‘
newspaper men Call "spot news", and therefore produced very little re-
action. The greatest criticism that I have on our own mark is our
inability toget into action and operation without considerable’delay."

Informal'publication has-had. a remarkable growth in the last
ten years. I Probably close'to a hundred series "of informal mimeographed
reports‘of a. periodic nature are being published by the‘Bureau of Agri-
cultural Economics. _ Many of the states have one or more regular series
of informal reports issued in mimeogr'aphed form. In addition to this,
many miscellaneous reports upon the results of specific projects are
being published from time to time. Most of these are in the naturp Of
progress reports detailing the findings from a given limited protion of
a major project. Many of them are sumaries of a year's work upon a
given project which contain essential material which may be of timer
interest to county agents, farmers and others immediately connected
with the farm problem. News articles is of course another way of
releasing such material. '

As. to the use of mimeographed‘ releases, one department head
writes:' ' ‘ ' '

"I favor considerable use of mimeographed preliminary reports
for interchange between workers and also as a means of making more def-
inite the findings of the Workers from year to year for their own files.
A great deal 'of material which we do not write up slips aWay into 013-"
scurity and is never used." , Another writes as follows: “Informal
publication is coming to play a greater and greater part in the pre—
sentation of the results of economic research. For this reason I
believe it important to encourage research workers to prepare brief
articles dealing with unpublished materials and results. Of course
the study should have proceeded far enough to insure that the results
given can be substantiated and the conclusions reached are sound.
Timeliness if often of great value and the cumbersome methods necessary
in getting out a. formal publication often delay publication until much
of its value is lost. Consequently, I believe the informal publication
should be given greater prominence than it has received” in theqast."

397

Utilization with A very large part of the utilization of

College Glasses the results of research in teaching is indirect
r .- or unconscious. This is especially the case

. , -- p when the same persons hemdle both teaching and

research. '- The analytical habit of mind required in the research is car-

ried over into the classroom work, making much more significalt the teach-

ing of the advanced courses especially. Those doing research in a field

are likely to be informed as‘to current developments in it, and the teach-

ing :that accompanies is sure to contain applications of principles and re—

search conclusions to the current situation. Research also gives to the

teacher a wealth of concrete information that is inValuable for illustra-
tive purposes, eSpecially in elementary classes.

It is true that teachers can have all of the foregoing and do
little or no research; but seldom is'thi-s the case. It is also necessary
to point out the danger that those engaged in research work become so
absorbed in it that they neglect preper preparation for the classroom.
Having teaching and research done at different times Will obviate most
of this. A fee may have to be taken off the teaching staff altogether.
The other danger is that the course material will consist too largely of
the results of research projects in hand.

,Where teaching and research are handled in the same depart-
ment, but by different persons, frequent conferences between individuals
may accomplish much of the foregoim. Calling upon the research worker
to present the results to classes not only gives this material to the
students, but brings the two staff members involved closely together.

It is entirely possible for a staff of six or eight or more
members carrying on several pieces of work along with teaching not to
keep closely enough in touch with each other. A number of departments
have been able to manage staff conferences succeSSfully, to which grad-
uate students are also invited, at which'results of research are pre-
sented, and problems of analysis are discussed as they arise. Current .
develoPments and new literature are also sometimes analyzed. Occasion—
ally men outside the department'or'the institution are asked to lead
discussions. It would seem that such conferences or seminars are
particularly advisable in the small to medium—sized departments which
have a number of young research Workers and part—time assistants.

The reports from institutions in'which research and teaching
in economies are in separate administrations indicate an effort, but a
far from adequate one, to bring the two lines of mrk together by ec-
casional conferences and visiting lectures.

(a) Bulletins for >
Class Use The medium which takes first place in direct

presentation of research material for class use is the technical bul—

letin. This use for technical bulletins is coming to be recognized

398
as sufficiently important alone to justify their publication. In a.
great many cases, this use can be combined with use by college-trained

farmers, farm journals and other equally eduCated people who have an-
interest in agriculture.

:- -" Next-in order is some discussion of the contents and organ-
ization of bulletins for class use. Following are a few observations
"growing out of experience; '

\

‘ 1. Such bulletins should not include elementary analysis
of the sort available in textbooks in economics, marketing
8116. the like.

2. They should.include, however a. careful statement of
whatever qualitative analysis is new to the project and
is basic to the quantitative work following it: also
reference to, and in some cases, a. summary of, necessary
qualitative analysis that is not readily available to

. students; for example, if available only in a foreign

language.

3. The foregoingwill mean in some cases the preparation
of technical bulletinsrin which the qualitative analysis

is more significant than the data,— perhaps the data merely
illustrate some qualitative analysis.

There is a disposition- for bulletin editors, or pub—
lications committees, to disapprove of bulletins largely
qualitative, or to ask that the qualitative portions of
bulletins be eleminated or greatly reduced. This reflects
their lack of familiarity with the social sciences. The
economics department may need. to educate them on this point.
In the past, however, the economics department when thus
treated4has more often than not been trying to include more
elementary discussion-of principles as research material.

4. 'A great deal of_-descriptive material is needed for
teaching purposes, especially in elementary courses. In-
structors should not‘have to sepdn their time in class
presenting facts and data. These should be available so
' that the students can get them themselves by reading.
Some of them are mrth having students take notes upon
in the reading, but for the most part, having them avail-
able for reading and later reference without copying is to
be preferred. It is‘highly desirable that the limited
class time available'be spent interpreting descriptive
material and data and in developing principles by induc-
tive methods from these. The classroom discussion should
be much more in terms of ghy than what; and the what questions
should relate to appliCati-on of principles and inferences.

399

The question ‘Lvill be raised at once as to whether as—
sembling; facts for class'use is research. Some of it
surely is not, and should be left to the teaching staff.
This statement surely applies to the mere assembling of
data from census volumes, or from surveys made at other
stations. But that which is collected by surveys from
direct sources is surely research material, but as al—'
ready pointed out, perhaps of a low order. The redeeming
feature is that the survey itself need not - in fact, ought
not - stOp with mere description; and any report of the
survey needs to combine some presentation of descriptive
material with the analysis of it - probably enough for
most teaching purposes.

It is highly pertinent to remark here that a fact
may be only a fact, and have no value outside of itself.
A Vast amount of almost valueless descriptive material
is constantly being inflicted on students. Primitive
curiosity about facts is constantly leading even our
best scholars astray. One would expect economists to
be especially careful to distinguish the relative values
of different clasaes of facts and save students from
those that lead to no action or are about as well for-
gotten as remembered. Some facts have value only to a,
few who are specially concerned, or when some particular
problem arises. Knowing, how to locate many facts is far
more important than knowing them.

5. There is a question as to how detailed an interpreta-
tion of data should be included in bulletins for teaching
use. There is no doubt need for some data to be pub-
lished with little or no interpretation so that the
students can practice their own powers of interpretation
upon them. This need can be met partly at least by
publishing; significant portions of the source data in
appendices. Teachers and students will always be able
to find other analysis to which such data can be put.
Students can review the data to see if the author's
conclusions seem warranted. For the most part, no
doubt, the need above described will be met at each
station by mimeographing such data.

6. Particularly until the time when a much larger
number of adequate treatises and textbooks is available,
and even after that, because treatises and textbooks are
seldom in the vanguard of progress, we shall have need
of bulletins of the sort that constitute essentially the
text for courses. This means that they must be much
more than collections of data; that they must recognize
that they are pushing forward the body of qualitative
and quantitative principles and statements that make up

ieconmnic science, and contributing that Which will
‘jlater be Forked into treatises and textbooks; In
' the newer. fields of agricultural economic research -
II " .Iq'rédit. and taxation, for example, the foregoing need
‘ is Ibest illustrated”

7. From'thé standpoint "of needs of teaching, technical
'bulletins' should include all explanations of methods
{used that is net available in textbooks, and even when
the methods" are entirely conventional, should explain
ItheI details of application to the particular problem.

For example, the units and measures and the Variables
used should be defined. If there are any aesumptions

IIIiIxnplicIit in the analysis, such as no change in the

demand schedule or income Varying with prices, these
" ‘ should be very carefully stated. There cannot be too

‘ 'much care given to vigorous thinking in material which

is laid‘ before students.

The foregoing particularly applies to bulletins
that are likely to be usedII in courses in statistics
and research methods. '

8‘. It is hardly possible to set before students too
many bulletins which are in proper scientific form,
WhiCh show on the face of them that the proJeCt has
been approached and‘ carried through in scientific "
order. Scientific thinking is a habit, and is
learned by conscious repetition at the start.

(b) Scientific ,

Articles The very best of the new developments in
agricultural economics should at once be published. in economic
journals so as to give‘them wide Circulation among co—workers.
Teachers can then bring them to the attention of their students
in classroom discussion, and graduate students can be asked to
read them. The best technical bulletins in agriculutral economics
fail to reach half the persons who have need for them. The scienti-
fic standing of agricultural bulletins is such as yet that busy peo-
ple do not know what they sometimes miss when they pass over announce-
ments of them hurriedly. Even the Journal of Farm Economics is not
enough of a recommendation for the most important developments in the
field.

(c) 3911.21:

Exercises The best practice in most cases is to duplicate
data from surveys and the like. Each institution needs its own, and
the material needs to be fresh. '

' i 401

(d) Slides V Excellent use can be made of slides for presenting
charts and even tables of material which is likely to be useful over a
period ~ such as curves of price movements, statistical demand curves, .
exPorts and imports, localization of production. But it is easy to
stock up with slides that are seldom if ever used. Bulletins and
books soon publish the best of this material. The blackboard serves
very well for a year or two, and by that time some better material may
be aVailable. The newer machines for throyving illustrations on a
screen directly from the pages of books are proving; very useful.

(a) Mimeorraphing Material. for courses not yet avr ilable in books
can be assembled by getting a supply of bulletins together in the
library; sometimes by having each class buy or send for the newer
ones. Still other mterial can be mimeogr..phed. The collection
of mimeograph material may be increased by special sections written
by the department, and eventually it may be Wise to prepare a book.

 

(f) Books Most persOns who are Working up a course have

a strong; disposition to feel that none of the books available meets

the situation,'£uid that they have in mind exactly the contents that

are needed. It reflects, of course, their ideas as to emphasis,

based on their own past experience and training. It may be a very

wrong emphasis; but it does not so appear to them. Some reason more
potent than a different idea as to centent should ordinarily be re—
quired before a new book is launched. There Lust be considerable

fresh materiel and analysis, and the prospective author should be one .
who has developed considerable of this new analysis himself. Ordinarily
he should have published several bu letins or articles in the field
before he undertakes a book. Agricultural economics needs more treatises
and textbooks; but they should not be written till the material is ready
for them. Books are being published rapidly at present in the field of
economies. Many new publishers have come into the field and they are
grabbing manuscripts recklessly. Leng~establishod houses are in some
cases taking manuscripts to keep their new competitors from getting

them. '

Young soonomists should realize that their first book
largely sets their colleagues—3's measure of ther: for all time. TWO
books well done mean vastly more in a professional life than a down
ordinary ones. It is said of the wrf- ter of a few good books that he
is a very able person; of the writer of the dozen ordinary ones that
"he is a hard worker, but of course a second-rater",

Most of the writers on the utiliZation of research material
in teaching stressed its great importance. They look rpon the teach—
ing as developing the future county agents, agricultural teachers,
mamgers of cooperatives and farm leaders, and emphasize the importance
of giving them sound training: in coonomics in their college days.

402,

One of the stations reporting on the subject of utiliza—
tion of research results took the trouble to inquire into the use
of their bulletin material by other departments in thicollegee. No
doubt the reports obtained were typical. They indicated that the
natural science departments are glad to have data and analysis for
conditions in their states. They like to have price data. assembled
also. Frequently they make larger use of them in the classroom.
They have wanted economic material to use in their courses, but have
not known how to get it or more. On the other hand, these other
departments did not think that the conclusions were always exactly
adequate, overlooking as they did important technical points. rJ.‘h:|.s

is further evidence as to the need of joint work on research pro-
,jects.

This raises the related teaching, problem of how to give
students the preper combination of economic and technical training,
best illustrated by marketing. Assuming that studies have been
published covering both the economic and technical aspects of
marketing of potatoes, should the material be reviewed in courses
offered in one or the other department, or in both? or how? All
possible arrangements exist somewhere. Those which make it needful
for each to present its own phases of the subject and those only,
are working out best.

403

PART FIVE

Tin; QRGAMUJIIJH JJ‘ .fldCDLI‘LfiJL, LOLCi‘tIt-S LESL'iARJH

If a division of agricultural economics in an agricultural
college or other institution is to function adequately and fulfil its
obligations as part of the institution of which it is a part, those in
charge of its destiny must give thought to a rumber of important prob—
lems of policy, relation to other subject-matter departments, relation
to other functional departments, and cooperation with outside agencies.
It is the pumose of this section of the handbook to discuss such prob-
lems of this from the standpoint of the agricultural economist in charge
of research in his field.

Vievmoints in . . If the agricvltirnl economist in charge
Econcmig of research is to think clonrly on the fore-
Research going, subjects, he must first of {ill have in

mind the different strndroints from thic‘n

economic analysis may be spurouched. First
is the distinction between he pure and the supplied, n-distinction that
runs across the whole field of science. Pure science or. the one hand
is taken to mean a body of L:cnerulizations '18 to the relationships be—
tween things without primary concern as to whether the 'lmowing; of these
relationships is of a y value to Eibeyd, as distinguished. from that
kind of science in which the problems attached are those whose solutions
will load to human ends desired of themselves. From another angle,
pure science is more abstract and more universalized then applied
science, which to quote marsh 1, ”deals with narrower questions moro
in detail"... mess the lines of distinction only partly coincide.
The point of- View of science for Science's sake may be carried over
into the details of special fields. It is being: done so increasingly
these days, and many fail to recognize that it may still be pure science.
In general it must be admitted, however, that those working within the
confines of mecial fields are mainly seeking for direct human ends.

There are differences to be noted in approaches to problems

of serving human ends. One appronch may be to find out that practices
and procedures commonly seen to attain these ends and then prescribe
them. Another is to discover the relationships between things from
which adentageous or disrxdvnntL eous results ensue. The first is
empirical; it provides the material for trade—school education. The
second is thorougdily scientific in spite of its objectives. The first
as a scheme of education produces a men who gets along; well as long
as nothing; changes. The second produces the chauzge~makerm Economics
islone of the fields in which the first kind of on aprroach is most
likely to lead to evil—claysw ‘ Maladjustzientn, which have been the

404

theme of more of the agricultural economic discussions than anything
else in the past seven years, are what results when people act with-
out understanding. ”The man who knows how will alivays have a job;
the man who knows why will be boss."

The distinction between social and individual must also be
kept clear. It cuts acrOSS both pure and applied science, and across
both of the applied approaches. The pure science of individual econ-
omics relates to the economic behavior of individuals; the pure science
of social economics to the economic behavior of groups. How the eco-
nomic behavior of the individual is conditioned by the economic group
of Which he is a part, lies in the border zone between these two.

The social or individual point of view in applied science
becomes a matter of objectives or ends. In individual or entrepreneur
economics, the objectives are priVate - profit-seeking and the like;
in the social aSpects of the applied science of economics, the objectives
are the larger ends of humanity - that is, human welfare economics. Very
frequently indeed do these ends come into conflict with private ends.

Boonomics is usually described as a social science. As a
matter of fact, in strict logic it is a social science in the same way
and to the same extent that psychology is. Both have an individual
as well as a social aspect.

Saying that economics is mostly Social does not mean, however,
that it does not concern itself with individual ends such as profit-
seeking. It is social just as social psychology is social, or social
history. Private profit-seeking is socially organized to a high degree
and socially conditioned at every turn.

Some economists have in mind, when they call economics a
social science, that it concerns itself with social ends as distinguished
from priVate ends. They are thinking of social in the sense of social
objectives. They are in the field of applied economics; yet these same
persons are likely to shy at applied economics. They fail to realize
that economics directed to social ends - welfare economics — is as much
applied as entrepreneur economics. Political economy, strictly defined
in the sense of state or public economy, is as much applied economics
as is farm management.

The distinction between social and national ends is also
important in times and places. It is very easy for the two to come
into conflict with each other.

A familiar but important contrast in applied economics is
that between short—time and long-time objectives. The latter are so
difficult to introduce into analysis that they commonly receive too
little consideration.

" 405

National and social may either both come into conflict with
various regional or lQCal o ojectives. Eerriment stations easily
drift into thinking in terms of purely local objectives.

tudies may be regions M1 however, and still be highly in—
spirited with national or social objectives. The great social or
national ends miist finally oe Worked out in tonne of local conditions,
and research looking to such local realization of ends may have a
higher order of social objectivity.

Just because a study is local in its habitat, it is not
necessarily local in its airs A study of land Utilization in a
limited area may develop principles or methods or analysis that have
wide usefulness or appliCation.

mhe Loc caliz Vation Secondly, the pers son in charge of a

of_ stearch program of agricultural economic researdh
Problems : should do some thin12ing as to the geographic

distribution of research problems and of

-ageneies and facili ties for working upon
them. There is no particular reason that the two should coincide.
AS a matter of fact, they do not. wiany pzoblems are cor;mbn to a
whole group of states, with perhaps, however,n finportant local differ-
ences. For exanple, severe? adjoining state 3 may grow hard winter
wheat, but the other crepe grocn tdth it may be dii‘ferent A state
may think that some problem connected with wheat is suificiently in»
portant to it to investigate it thoroughly. But if a neighboring
state does the Same, Hmch of the work will be done twice. The local
flfl:ifferorces that appear when several states work tog ether on a regional
prob lem often are helpfil to an understanding of'the whole. In some
cases, only a small area is involved, but it is split between two
ssates. ' ‘

. Another situation'is that in which several states dispose of
a product in one consuming center L for.example, milk in the Chicago
market; or at the sane time in several competing markets, as in New
England.

The marketing system for a commodity like cotton reaches into
great national and international markets. While some important national
markets may be localized in one state, they are also of interest to other
stateS. In other oases, livestock, for example, a whole set of inter—
related central mnkets is involved. A state like California may be in-
terested in some foreign markets that no other state is; but this is not
the usual situation. \

Oftentimes it is extremely desirable that the production and
the consumption be analyzed in relation to each other - and they may be
at opposite sides of the continent, or even in different countries.

406

There are many problems common to all states, but of such a
nature that they need to be studied locally - for example, rural living--
One state may have a dozen sets of local rural living conditions each
of which needs to be analyZed by itself. But it will be highly unfor-
tunate if these loCal studies in different states are not made on such
a basis that some condusions and principles can be derived as to rural
living in general. Taxation and credit illustrate this same thing.

In many of these, questions of broad national policy are
also involved, making it advisable that they be considered in their
entirety. some have to be considered from this point of view almost
entirely - for example, the tariff and immigration.

Lastly come the problems in pure economics, both general and

special, and in developing methods of analysis, which are mostly with—
out habitat.

W The outstanding features of the local-
W iZation of facilities for handling research
Facilities. problems are the following:

1. Local experiment stations find it ex-
pensive to collect data in distant markets,
and. are almost necessarily poorly informed
as to .the problems of such mrkets. If only one such market is involved,
these difficulties are not so serious.

2. If local stations are to undertake many problems, they need
data on a large scale or over a wide area which it is difficult or ex-
pensive for them to obtain.

3. Local stations are constantly under presaure to undertake local
and immediate problems, and those with individual as distinghished frnm
social objectives. Extensive work is necessarily loCal in large measure,
and more than present available resources are needed to conduct the re—
search which is basic to the local extension and teaching programs.

4. Local stations are usually not adequately enougi supported to
undertake expensive research in problems requiring attack on a large
scale or on a wide basis.

5. Relatively few research workers in lOCal stations are broadly
enough trained to handle many types of problems.

6. Local research staffs are sometimes not strong enough to
handle even the local asPects of some regional problems.

7. In some cases stations located close to central markets or
inmortant consuming centers have staffs capable of undertaking this
end of problems reaching into them; but never enough of a staff to

:"-.4
1.0

407

handle all such problems. The state authorities do not ordinarily
appreciate opportunities of this kind and develop their departments
of agricultural economics as they migit.

_8. There is therefore great need for research agencies which
can operate over a wide area and on a larger scale; whiCh can also
undertake broad national and international problems of agriculture.

9. Since many problems require joint effort of several stations,

there is need forsome outside coordinating agency and also for leader—
ship.

It is clear that the United States Department of Agriculture is
the agency best situated to supplement the work of the states in most
of the respects above noted. ‘

There are several reasons, however, why the U. S. Department of
Agriculture alone does not suffice. Toese are as follows:

. 1. It is far from being sufficiently supported to do all this
work. ‘

2. In problems of broad economic interest, it is highly desirable
that several different agencies work more or less independently in order
that all approaches and points of view receive consideration. '

3. There- is danger that a department of the Government will get
the reputation of representing the point of view of the interests it
represents and have its findings discounted in consequence. Too often
such departments actually deserve such a reputation. There is need
that some agency should be "attorney for labor," or "attorney for
agriculture;" but it should not be a national research agency.

4. The U. S. Departmentof Agricultural is constantly under
pressure for results that will be imediately helpful to some group. '
Under these circumstances, it work tends to lose much of its scien-
tific quality and to become empirical. Problems of fundamental re-
search valus tend to be pushed into the background. This is more true
in the field of economics than in perhaps any other field.

5- The policy of the U. S. Department of Agriculture thus far
has been to insist upon men for its economic staff who know the field
and the commodity in 8- practical way; also who can establish good con-
tactS. One does not want to decry the usefulness of such qualificap'
tions for a large part of its present work; but if it is to assume as .
one of its Paramount obligations also the developing of economic science,
then it met select some of its staff primarily because of scientific
training and ability. Unitl such a time as it changes its policy
in this reapect - perhaps it ought not to change it .. there will be

4'08

great need of other agencies doing research work on more fundamental
lines than it can pursue.

The problem of division of the field between agencies and of
cooperation between them is to be considered in a later section.

Research The division of agricultural economics
M which has thought through its problem from
m: the two foregoing standpoints is in position

to formulate'its research"policy and program.

_ ,. The statement of policy which anew agency ’
formulates at the start of its operations is bound to be highly tentative;
but still it should be made. Frequent inventories of policy and program
should be made at later stages, so that any departures taken from former
plans will be recognized and analyzed as to their desirability. Changes
in personnel, in financial support, or in economic conditions, may often
make changes necesSary in the policy and objectives of research agencies
in the field of economics, but ,all important changes should be consciously
made and for sufficient reasons. A wavering uncertain research program
in any field is extremely wasteful of human energy. The broad general
alternatives in the matter of policy which are Open to a research agency,
are the following:

1. Development of the science of economics 15 attacking Specific problems
of amelioration.

2. In the latter case, attacking local problems, as within a state, as
against attacking sectional or national or international prob—
lems. The first of these means essentially, in the case of
state experiment station, the working out‘of‘a state research
program looking directly to improving conditions especially
within the state. ‘ ' '

3. Research in private v_s_ public economics - in problems of adminis-
tration of privage enterprises E those of public agencies.

4. Current and temporary problems lg those of long standing and.
likely to "be persistent.

5. Controversial issues of the day, such as the McNery—Hausen plan.
YE. problems of a more general nature.

6. A largely opportunist policy, choosing-projects in response to
public interest in, them, or to suggestions from without, :15
a long-time program based upon a careful analysis of the
needs of the field.

7. Cooperation with other research agencies in larger projects
15 independent work on projects not requiring such co-
operation.

“09

8. A policy looking to research results only 15: one which develops
research workers in addition. '

9. A policy which coordinates research with the needs of the teaching
and extension program of the institution E one which considers
research objectives only. '

10. Whether little or much attention will be given to improving re-
search methods. ' '

Obviously theseobjectives are not all mutually exclusive.

A project designed primarily to meet a situation in one section of a
state may at the same time'illustrate and clarify the application of
‘ an important principle of economica,- for e'Xample', it may shew how
the principle of comparative advantage works'out undercondi'tions of
lagging readjustmentsgwor it may developla hitherto “unused method of
measuring lag. Manyl'states often have the'same problem, and a solu-
tion for. one state "is of great service to the rest. ' ' '

Even '.though some of the foregoing objectives are rznitually
exclusive, this does not prevent more than one of them being included
in one policy.'.' For example, a program‘may include projects in the
fields of both private and public ecdnbmi'cs: or both'lo‘c‘al and national
problems.

Neither does 81 Politiynneed-to ,be. altogether systematic or
altogether Opportunist. It can be the £0 rme-r- and yet sufficiently
elastic to provide for important emergencies, .,

The policy which any division should adopt depends upon many
things. A division in a new area with many different problems of
developing a profitable, system of agriculture for. its state, can be
largely excused if for a time it gives little attention to the needs
of the science of esonomics. A division in a small state should
properly consider itself as having an obligation to devote some por—
tion of its federal supportrvto the larger regional'problems or to
developing the science 'of. economics and its methodology. The more
ample the support which the ivork in economics reCeives, other things
being the same, the more attention should be given to the latter sort
of problems. Divisions cf agricultural economics‘so located or
developed as to have'attracted' strong groups of graduate students
should plan their research programs fully as much with a view to
training research workers as to obtaining direct research results.

No condemnation can be too severe of an institution which is so

narrow or blind that it cannot see the opportunity for service that -.~~\
such a. situation provides, and refuses to support 'its department of _
agricultural economics on this basis. Institutions with departments
made up mostly of younger men without research experience should in

41.0 :13.

general work upon narrower and more local problems. The same is
equally true if the staff is largely made up of older men without
special training in economics and statistics. A department with
meagre support may find it best to concentrate on more local and
immediate problems until such time as, following such a policy, it
secures more ample support. On the otherOhandLif such a department
has a staff of high-grade men well trained in economics and statistics,

its forte may be to Concentrate on projects of a more fundamental nature,
and neglect somewhat the immediate service needs of the local situation.

With the foregoing by the way of introduction, let us con—
sider some of the above alternatives in more detail. The science of
economicshas reached a stage where‘caref‘ul detailed work with the be—
havior of its principles under the greatly Varying conditions of time
and place, field of production and Sphere of consumption, and institu—
tions is its urgent need. Except in the sphere of cbrnamics, it is not
probable that many really new principles will be stated; but nearly all
of them need to be‘stated more explicitly, and also more particularis—
tically for different situations and conditions. Only when so stated
are they especially useful for those who have tasks of administration
either in the public or the priVate field. The present is an era of
"applied" economics, using the term a plied as Marshall used it in his
footnote on page 37 of his 1910. edition:

1"”

"Some parts of economics are relatively abstract or
pure, because they are concerned mainly with broad.

. general prOpositions: for, in order that a proposi-
tion may be of broad application it must necessarily
contain few details: it cannot adapt itself to par—
ticular cases; and if it points to any prediction,
that must be governed by a strong conditioning clause.
in which a very large waning is gven to the phrase
"other things being equal. " ’
"Other parts are relatively applied, because they

deal with narrower questions more indetail; they

take more account of local and temporary elements;

and they consider soonomic conditions in fuller and
closer relation, to'other conditionsof life.",

In a word, then, reducing principles of economics to a
working basis is the major task of the moment so .far as the science
of economics'is concerned; and agricultural economics is in the posi-
tion where it must make a larger contribution along these lines than
any other ’division of applied economics, because the conditions or
combinations of conditions and the situations it presents, are more,"
likely than the rest to be somewhat unique, Or to place an unusual
emphasis upon certain aspects of the principles. At the same time,
the Whole organization of research in :5in cultural economics in this
Country is especially favorable for developments along the foregoing

411

lines. The resultsdesired are to be slowly formulated out of a
large number-of studies organized on the same area or commodity or
period—of-years basis with which we are already so familiar. To
make these contribute to the ends desired, it is only necessary to
give many of them a more scientific cast and methodology, and more
of the points of vietr and objectives of the higher orders of re-
search. Almost any such study, if planned and Bupervi sed by a
man with graps of economic relationships and command of research
technique, can be brought sufficiently Within the sphere of econ-
omics as a science tO'add'a little to its progress.

But the objective of developing the science of ecanomic's
should not be entirely subordinated to the position of being inci— »
dental to the more immediate and popular ends. Projects should be
deveIOped whose aim is principally to discover the workings of
Specific economic principles, or to'assist in a more explicit form-
ulation of them, or a formulation more particularly in terms of agri—
cultural conditions. The data examined will of course relate to a'
particular period or area or commodity, and other results will be
obtained which will be of service to those persons or agencies
Specially concerned with this period or area or commodity; but this”
service is essentially a by—product.

"A similar statement can be made with respect to methodology
as a research objective. A study vinich slavishly accepts the units
and measures and forms of analysis of earlier projects contributes
little if anything to progress in methodology. No doubt there has
been more of this in the past than has been desirable for the long-
run usefulness of the science of economics. It has probably been
necessary because of'the‘newness of the field and the scarcity of
men broadly enough trained to be able safely to prosecute independent
modes of research. Pressure for immediate results of use to local
interests has been another factor. An unwise over—emphasis on stand-
ardiZation in research method has been another factor. Standardiza—
tion in research is dangerous at any stage. It is particularly _
dangerous in‘ the early stages of growth, checking all progreSs and
producing a stunted dwarf science. A handbook such'as this one
might easily be, could do vast injury to agricultural economics by
presenting definite rules cf precedure. Those responsible for it
have had this though consciously before them at all times. Their
wish is only to summarize the experience in methodology to the
present, and present this as a foundation upon which later progress
may be more effectively reared. '

In contrast to the foregoing, almost any project in econ—5
omic research can be so planned and executed that it adds something
of value to research experience; It may be nothing more than further
evidence as to the accuracy of survey answers obtained to a Egrtain

4ié H,

type of question. ' Even this is in contrast with the too freuuent
practice of assuming that because others have accepted answers to the
type of question as valid, or even have proved them to be valid for
their study, they must be valid for the present study. Those in
charge of this‘handbook have been trying to get together the results
of experience with all sorts of research procedures, but they have
met with disappointing results. It is very clear to them that few
indeed of those who have prosecuted studies have sufficiently tested
the validity of their results, or have sufficiently analyzed their
project from a methodology point of view to see what light it “throne.
on methodolog in later projects of the same general type. Or even
if they have, they have not thought it worth while to reduce their
conclusions on the subject to definite form and publish them. In
consequence, much of the value 'of many studies has been lost. A’
majority of the projects of the future'should be set up with a view
to making contributions to methodology as well as those of the more
usual sort, and the’ results in this sphere should be published when—
ever possible, even if as nothing more than notes of a page or tm
in journals. Needless to state, such material is of great value in
the same institutions in courses in statistics and resaarch method.

In addition to the foregoing, projects should be set up
whose objective is primarily the development or testing of research
methods. . An example of this is U. '5. D- A. NO. 1277, "Input as
Related to Output in Farm Organization and Cost of Production Studies",
Some of the price analysis projects have been ofthe same nature.

So far as research devoted to serve local needs is concerned,
the trend today is strongly tOWard relating it to a definite long-time
program of continuing adjustment to changing economic conditions as
set against the background of the physical environment. ll‘his has
meant, and wisely in most cases, a more careful survey and analysis
than had previously been made of the physical facts of climate, soil,
topography, systems of farming and, the like as a basis for later and
more essentially economic analysis. It also means setting up machin-
ery or making such other arrangements as necessary. for securing the
data of economic change, of acreage, production and yield of crops,
of prices of farm products, wages, prices .of machinery and supplies,
amounts of labor, fertilizer and other supplies used, of market ‘re-
ceipts and shipments, taxes, land prices, rents, etc. , For all of
these changes, suitable measures must be established and kept up—
to—date. Out of an environment thus analyzed, it is easy to select
that area or commodity where special studies in more detail can be
most adVantageously executed.

Carrying through such a, program as the foregoing may be
more of an undertaking'thsn is possible with the financial support
provided in some states. It surely is not more then should be
possible if available state and federal funds are properly appor—

4:13

tioned as between the soonomic and other work. If the funds are not
available, the best procedure is probably to mark off a pcrtion of
the field — certain commodities or certain areas or certain divisions
of the field of ecanoufics - for special analysis. Under such cir-
cumstances, many states will do well to omit projects in transporta—
tion, public finance, Credit, land tenure, etc., from their research
program.

Research in matters of national policy, in public adminis—
tration as it relates to agriculuire, in taxation, railway rates and
the like, can ordinarily be handled best by those institutions in
which the department of agricultural economics ozr cooperate with a
general department- of econonlce In the Same institution... lic-E‘iciencies
in this respect my be partly overcome by working closely wi th the
U. S. Bureau of Agricultural Economics. There are largely local as-
pects of some of these problems — taxation, for example — which can be
has: led effectively on this latter basis. Also any division of agri-
cultural economics may have some member of its staff who is Specially
trained in one of these fields.

If it is the duty of state agricultural institutions to
aegist with the handling of temporary problems or take a hand in the
controversial issues of the day, then it is the function of the re-
search staff in soonomics to c-ollcct and analyze whatever date. are
aVailable and pertinent. Work of this kind should not be turned
over to extension specialist, but should be done-in close coopera—
tion with them. If accepting such a responsibility as the foregoing
takes an unduly large part of the time and energy of the research
staff in eCOnomics, then the institution is ei titer concerning itself
too much with such problems; or it is not supporting its department
of agricultural soonomics sufficiently. Most heads of departments
of agricultural economics will. testify that if they concerned them~
selves with a third of the current and controversial issues that
arise, the whole energy of their staff would lie-absorbed by it, and
there would be little left for teaching and more vital research.

‘The policy of training research workers in conjunction
with actual research has serious limitations as well as advantages.-
Carried very far, it involves a tremendous burden upon those in
charge of the project. But where one trained and competent leader
is in charge of one, or at the most possibly three projects, these
not running in the same phases contemporaneously, So that he can
give close supervision to the details of each, then the two func—
tions can often be combined effectively. The personal. equation
is an important factor in such a question. Some othonwi se compe-
tent research workers have single-track minds and can think of only
one project at a time, and follow closely nobody else‘s work except

'41};

their own. Others benefit from directing more or less parallel 8tUdieB'
Research must be learned in large part by doing. In many fields of
StUdY, about all that is needed is to turn a graduate student loose in
a library and let him work out his own salvation. The research in
economics that promises results these days is mostly with data. Often
these data must be collected at great expense. The necessary calCUl8~
tions with the data are often so laborious that graduate students can
perform only a small part of them except at a great waste of their
precious time in school. Training graduate students for their work
with research in economics therefore requires either that they be set
at work on some practice exercise, or that they be closely supervised
on real problems with real data. The former is suited to formal
courses in statistical method; the latter to research semdnars and to
theses. Any department of agricultural economics which expects to
given men their final training for research must work out the technique
of combining research training with actual research work. Experience

in collecting data is the easiest to provide. It is with the analysis
that the real difficulty comes.

So much in need of improvement are the curricula in agricul#
tural economics these days that it may well become a major objective
of the research work of some departments for a time to develop material
to be used in certain courses and ultimately in text books. Since the
manor emphasis in teaching economies should always be on principles and
analySis, Such an objective need not interfere in the least with research
aims of a high order. If the teaching ideal of the department, however,
is mainly to supply the students with data, a different statement might
be needed.

No discussion should be needed as to the desirability of a
research program that provides the basis for a broad fundamental exten-
sion program, both based on the sane careful outlook analysis.

~91§§§l£iggfiign A division of agricultural economics
93;§§§§§§E§ , also needs to have clearly in ndnd some
REleggfii classification of its field as a basis of

organization of its research program, The
classification which the Advisory Ccmmittee
has adepted or its analysis of scope and method, is the following, which
is in terms of the fields of agricultural economics as these stand forth
in th9.minds 0f workers 15 the field and reflect themselves in the names
0f 05133636818316. researc}; projects:
. . .

f'Oihrm Management-'

.Market (including cooperative marketing)
9 V

. Prices - -,

I : s

415

Land Economics

Consumption Economics (including rural living)

‘ Transportation (including roads).

Credit

Insurance

Taxation

Rural Population

Commodity Studies (combining production, marketing prices, etc.)

Geographical Studies (including area studies combining produc-
tion, marketing, etc.) 1

Historical Studies
Agricultural Income

National Agricultural Economy (including esPecially national
production programs, the tariff in relation to agriculture,
relation of agriculture to industry, immigration.)

Outlook Studies (foregoing focused on the problem of fore-
casting impending developments.)

It is assumed that most of the projects relating to market
prices will be included under Prices. Consumption Economics includes
research in factors influencing consumption as well as the consumption
of rural people themselves. It is realized that some aspects of rural
living will more properly belong under Rural Sociology. The same is
true of Rural Population. The category Commodity Studies has been
been included to take Care of projects organized at- some stations in
which all aspects of one comedity are studied. Geographical Studies
include,among other things, area. studies, in which the area. instead of
the commodity is the base of organization; National Agriculmral
Economy of course includes many aspects of other classifications, but
as used here is meant to provide a'place for projects of this nature
not otherwise taken care of.

Research Outlines .is soon as possible after research
by Fields. work is undertaken in any one of the fore-

going fields, a research outline should be
prepared for it to serve as a guide in
' mapping out projects and keeping them in
proper relation to each other. As a guide to making such an outline,
there is here submitted a sort of generalized outline for one field,
that of farm management:

416

I Description of farming in a given areas, or of the pro-

II

III

duction of a given commodity. The description of farm-
ing indicates the products produced and in what amounts
and proportions, and the methods used in producing them.
Sometimes called "type of farming" studies.

Determination of what products to produce, and in, what
proportions, in order to secure the largest net income.
Also sometimes called "type of farming" studies, the
difference from the foregoing being that here the ob-
jective is not merely to describe the type of farming,
but to determine which type of farming pays best so far
as combinations of products is concerned.

Projects under this head should include types and
qualities of each product as well as different species
of product - e.g. , Grade"A" vs Grade "B" milk.

Determination of how to produce each product at lowest
cost Consistent with highest net income to the farm as

a whole. Projects under this head are Variously

named. Many are called projects in "farm practices".

It will be impossible to give an outline of them that

is either Complete or entirely logical. Let the follow—
ing be interpreted rather as a list of types of projects
coming under this head:

A. Those relating to choice of the cost factors -

1. Whether to use'tractors or horses, combines
or binders, etc.

2. Whether to harvest with amachine or with
man—labor.

3. What type of machine, horse—labor, men-
labor, etc., to use.

4. What Varieties of a crop, or breed of
' livestock, to use .to turn out a given
type of product.

18. Those relating to choice of Operations: disking
Igplowing for small grain; hogging off y__s_
husking corn; stacking YE. shock~threshing.'

0. Those relatingto time of operations: when to
plant, when, to fatten,_' etc.

D. Those relating to the quantities of various
factors, such as labor, seed, feed or fertile

417
izer, to use. Applied'to labor, such projects
take the form of determining‘how many times to .
milk per day, how many times to cultivate the
corn, to spray the apple trees, etc. Such

projects have recently been called "input-
output" projects.

IV. Determining the most advantageous size of business, or.
s00pe of. Operations, for farmers of various capacities,
and in general under a given set of economic conditions.

V' 1 Determining all of the foregoing with respect to a 1 _.
. particular product, usually called "enterprise'fstudies",a
"or, "comodity studies" , although the latter usually 'in— ‘

clude more than theproduction aspects only; '_'

VI I Regional or area. studies - in which all the foregoing
‘ ' or a considerable part of them, are determined for the
major types' of farms and products produced in a given

VII Outlook studies, in which attention is focussed upon .
the changes in all or some part of the foregoing which
are neceSSitated by changing economic conditions.

VIII Determining the most advantageous layout of the farm
and farmstead and farm buildings.

Ix Determining the best leasing arrangements.
Relationships Another important organization problem
between research which an economics division needs to con-
in economics and sider is the relationship of its subject to
in other 31330390135: ' ’ other subjects handled in the same instituo-
(a) the natural ‘ tion. The distinctionbetwem economics
sciences. and natural. science hasalreadybeen made.

It remains to. show how; research in economics
and the natural sciences, more particularly the applied natural sciences,
the cultural and husbandry courses,” are related, to each other in actual
research problems. PrOb‘ably the best way to discussnthi‘s-is to take a
number of cencrete'cases.~ Let jus'start With the so-called intensity
studies, in which costsr’and profits are related to the inputs of certain
factors and the resulting‘outputs. Research in this important field
seems to be held in abeyance at present because no understanding has
been reached as to the relationships involved. Clearly it is the task
of the natural science departments to conduct the eXperimental Work in
which output of milk is related to input of feed, or yield of cotton to
input of fertilizer, or amomt of irrigation water to yield of crops.

415

Clearly it is the work of the economics departments to collect and an-
alyze the necessary information as to prices of the input factors,
prices of the products, farm organization and relationships to-other"
crops and liVestock'enterprises, and to estimate the probable effect
on fan: receipts and expenditures and the farm business as a whole,

of different combinations of the input factors. Both sorts of an-
alysis are needed for a complete answer-to the question.

The curves of output per unit of varying inputs which are
needed for such analysis can be derived either experimentally or
statistically. EXperimental methods will give the results in purer
form. .Statistical method is not likely to be able to measure enough
of the varying factors, or measure them with sufficient precision, to
give very high accuracy to the statement of effect on'output.' On the
other hand, the statistical method, under reasonably favorable circum-
stances, cen include types of variables than cannot-be introduced into
an experiment, mhich exist under actual conditions of'production. The
results obtained may therefore more nearly fit actual conditions than
those obtained from controlled experiments.', Moreover, we must not
exaggerate the accuracy of many experimental results.. It is seldom
possible to control all the'conditions. Measuring-the remaining
factors does not help greatly so far as one experiment‘is concerned
because not enough cases are included to make it possible to isolate
the effect of differences in them. The results obtained can be
very accurate for the conditions under which the experiment was
conducted, but how they fit other conditions Cannot be determined
from the experiment. A statistical study of fertilizer effects on
a large area may take acCount-and measure the effect of varying
rainfall; an experimental study gives results with one given quantity
of rainfall. If the experiment is continued from year to year so
as to include different amounts‘of rainfall, then diseases and pests
or growing season may not be the same.~

From the foregoing it~appears that the experimental and the
statistical method are valuable complements of each other. The re-
sults will not always check;~but this merely becomes an occasion for
further work on the problems; 4 by the_experimentalists to see iflthe
special circumstances under whiCh their trial was conducted did not
give these particular results; by the statisticians to see if vari-
ables not included have not modified their results; or Whether thier
measures or data are not defective." To illustrate the latter, if
pasture feed is omitted from a_statiStical analysis of steer feeding,
the effect of rate of feeding grain'may be modified greatly by'the.
fact that the slower the rate'of feeding, the longer the steers run
on grass. ' ' ':” H ' " " . If. -

419

The economist needs his results from such studies in the
form of curves so that hecan locate the points of lowest—cost and
highestuprofit combination and so that he can estimate the effect
of different rates of input on gross and net farm income. The ex-
perimentalist can give him these results in the form of curves so
that he can make his sstgup accordingly. But this is Something
which the experimentalist has not been accustomed to do, and only
with difficulty in most cases can he be made to change his procedure.
It must be admitted also that it is not aIWays say for him to arrange
his set-up in the necessary form. But vastly more can be done in
this direction than has been done in the past. Also there is no
reason why the natural science departments should not use the statist—
iCal method, which gives the desired curves directly, as a complement
to the eXperimental method. Those that have a prejudice against the
statistical method must lose it and learn the necessary statistical
technique and its underlying philosophy. -In many cases, astatietical
attack on the problem may be the next step needed. '

In many cases, no doubt, the natural science departments will
prefer to confine themselve to the experimental method, and will gladly
have whatever statistical work is done on the problem handled by the
economics staff.- But the economics staff will make a failure if it
does not have the'assistance' of the natural science departments at
every turn.* - ~ ~

One might analyze in similar way the problem of combination
of enterprises. The effect of different crop rotations on yields
and soil maintenance are problems for the departments of soil, agronomy,
horticulture, etc. The economist must have the results of such studies
in order to complete his analysis of the effect of different combinations
on gross and net farm income.

All research in the economics of land utilization must lean
heavily on information as to soil and other physical conditions that .
must be accumulated by natural science departments. Similar analysis
is also needed as to .the relation between temperature, growing season,
rainfall, soil, etc.,_ and the growth and yield of the different crepe.
But the final responsibility for statements as to the agricultural
possibilities of .a section should rest with the economists, for only
they can have made sufficient .analysis of markets‘and trends in compe-
tition and prices to "hazard the forecast involved. It will not be
safe to have a soil specialist whose forte is peat soils make predic~
tion as to what is going to happen to the peat land. of his state in
the next fifty years. But he has the most important contribution of
any to make to the final answer to such a question. The problem of

 

* For further mderstanding of this type of researCh problem, see
Wisconsin ResearchBulletin No. ,79,"Practice Responsible for Varia-
tion in Physical Requirements and Economic Costs of uiilk Production
in Wisconsin Dairy Farms". ‘By iii. A. B. Ezekiel, P. E. Norvell and
F. B. Morrison.

480
land classification in the final stages is an economic problem.

“Marketing offers an abundance of problems in which economic
and natural science analysis are closely related. ‘lhe early research
of the U. S'. Department 'of Agriculture in the field of marketing was
in problems requiring natural science technique mostly — grading,
sorting, packing, shipping, storing, etc. Economic analysis of market-
ing problems was largely introduced when Henry C. Wallace becama .
Secretary of Agriculture. The present, staff of the Bureau of Agricul-
tural Economics engaged in marketing research includes both economists
and men who know their eommodities. I' The two need to work together
on many projects - grading, grade price differentials, marketing channels
and margins, problems of marketing business units, etc.

(b) Relationship 1 i " ‘r Agricultural economies has close rela—
to other social ’ ' tionships to political science and sociology.
sciences. , _ I , In fact, so close is the relationship that in

many institutions whatever research is done

in these two fields is‘ ”as part of economic
proJects. The task of administering the affairs of a township or county
is usually considered a concern of the political scientists. The polit-
ical theory of the function cf the state is basic to the whole pregram of
regulation and service in the interests of agriculture. But there are
aspects of it which are most certainly economics. I For example, whether
it is the best use of‘the social incomefof a township or county at a
given period to devote part of Iit to building a town hall or bridge or
improving a highway, or whether to use the funds in one .of these ways or
let the farmers have the money to pay off their mortgages, are essentially
economic questions. Assessment of farm real estate involves economic
analyses principally.

The relationship between agricultural economics and sociology
is best discussed when the term sociology is defined rigorously so as
to include only group analysis— group structure, group behavior and group
functioning. ‘ It then becomes a basis science like psychology upon which
applied economics draws for the solution of many of its problems. In turn,
sociology is able to draw upon the principles of economics for an explana—
tion of many things in the structure, behavior and funcioning of group8.
The migration of rural population to the city, for exemple, must be ex.-
plained in large part in terms of the economic principle of comparative
astntagefI The differences between the rural and the urban family
are partly based on economic factors. Exanples of economics making
use of sociological data and principles are found in- studies in farmers'
attitudes toward economic issues, membership relations in cooperative
organizations, farmers' movements, the economics of rural conswrption.

Much of what is included under the term "rural life" is more
largely economics then sociology. The content of rural living is
essentially an economic phenomenon, the result of the efficiency with

421

which human resources and the resources of nature are utilized in the
productive process on the one hand and in consumption on the other.
The essential idea of soonomics is utilization. The principles of
economics are the principles which explain how to utilize human and
natural resources so as to secure the maximum satisfaction of human
Wants from them, satisfaction being understood to be equivalent to
human welfare as Pigou uses the termin his "Economicsof Welfare".
There are many things. about rural life, however, which cannot be
understood without a proper understanding of group behavior and
functioning. The standard of living, for example, properly under—
stood as the content of living we strive for rather than that which
we sxperience,‘ must be explained largely on sociological grounds.
The sociologists as yet have made little study of standard chalking
as a proper sociological concept.

Much of the confusion as to the sphere of sociology comes
from an erroneous concept of certain values in living as being social
as distinguished from economic. Economics is a social science, and
the values it contemplates as ends are social values, that‘is, human
welfarevalues. Values in the purely sociological sense, if there
are any such, are values which relate to the functioning of the group,
as such. . ‘

Psychology stands in the same general relation to economics
as sociology. It provides the basic explanation of human behavior
which is needed for an analysis of many economic problems. *

 

(c) Relation to The esonomics of home economics as a
home economics field of teaching and research is about
research. where the economics of agriculture Was in

1910. Here and there departments. of home

economics in agricultural colleges are be-
ginning to realize fully that the household side of the farm family
economy has about the same range and types of economic problems as the
income-producing side of it - its problems of economic organization,
layout of work, financing, marketing (buying), etc. In some institu-
tions, the department of agricultural economics is taldng an active
part in developing the work along these lines; in others not. That
part of it which relates to the way in which the farm income is ad—
ministered as between farm and family uses, as between present and

future, as between major divisions of the budget, is in many institur- .

tions considered as economics of consumption as set over against the"
economics of production of farm products. ‘

 

* Davenport' 3 definition: "A level of consumption so fixed.that any
falling short is felt as a privation‘.

.....r'

422

The line of distinction between the economics of home economics
and the rest of the work in home economics is the same as was pointed out
for agricultural economics and the applied-natural sciences of agricul-s.
ture - the economics of home economics concerns itself vi th the value and
price aspects of things and the orgeniZation of productive resources of
the home in such a way as to secure maxinnm utilization of them. It
begins like agricultural economics with certain resources in Capital
goods, labor and current income, and aims to secure the best possible
utilization of them, values based on alternative uses and prides being
the Special desiderata. ‘ '

It 'will be apparent at once that many of the research problems
of the eccnomics of home economics require a combination of economics
and natural science analysis. The most conSpicuous examples of this
are in the buying side of the household economy.' Questions as to pro-
portions of the different kinds of foods to buy, what types of fresh
fruits and vegetables, vfnat sizes of oranges, what classes and grades
of eggs, what kinds and cutsof meat, what types of‘dress goods, or
what types of floor or wall coverings, require a high order of under-
standing the facts as to the physical or chemical properties of these
things and of the needs which they are to fill, as a basis for the
final economic choice, in which these considerations are set over
against price and the Various alternative uses of income.

Once studies in rural living get beyond the point of finding
out how farm peeple spend their incomes and attempt to develOp a better
understanding of the principles upon which such Spending should be
done, the analysis of the technically trained home economist will become
as necessary as that of the technical agriculturist in inputs-output
studies with feeds and fertilizers. Rural living studies should soon
pass out of the descriptive phases in which most of them are at present.

It is highly signifiCant that a very considerable proportion
of the Pumell projects in home economics related to problems in tech.—
nicel home economics. - Congress in passing the mall. Act assumed
that home economics Was ”economics. What they undoubtedly had in mind
was the problems of rural living in which economic considerations weigi
heavily. That the home economists have not seen-‘thematter in the same
light as Congress is clear evidence that they have narrowed their think-
ing down to the technical or natural science aspects of things, just as
the agriculturists had done in the long period from the establishment of
the first agricultural college until the change came perhaps ten or
fifteen years ago.

EQL'EM lbrestry has its economic problems no
forestry less than agriculture and of the same general
rgsearch order. But thus far they have received little

attention. There are few trained "forest econ-
omists" - men who have acquainted themselves with the technical aspects
of forestry and are also trained in economics. The outstanding problems

423"

of forestry in the United States at present are economic, of determining,
in view of the prospective prices of land, labor, capital and the other
cost factors.on“the..one hand, and the prospective prices of timber pro—
ducts on the other, what kinds of woodlot and forestry enterprises are .v
liker to be‘profitable in the next fifteen to one hundred years. There
is only a limited amount of information available that helps to answer
questions of this kind. Many of the leading thinkers in the field of
forestry are aWake to this situation, and funds for‘the expansion of
economic research work in forestry are being vigorously sought.

 

.Relation of Research It is equally necessary to have a clear

to other functions - set of ideas as to the relation of the re-
search to the other functions - teaching,

(a) To Teaching extension, etc. , It should go without ex-

planation that the teaching in any subject
can be no higier plane than the research. In anyone department it
may be on a higher plane as a result of making full use of the research
results of others; but not for all departments combined. Many teaching
departments are to be criticized for not following closely enough the
important researches in the field of economics and incorporating their
results and analyses in their courses. One of the reasons for this is
that‘the methods followed in the met significant studies are often be-
yond-the reach of many of the teachers. This will be remedied as more
and more of the teachers have adVanced training. Inadequate training
in economic theory prevents many teachers from giving a sufficient an—
al'ytical content to their ocurses. This is no doubt the greatest weak-
ness of the present teaching in agricultural economics. That agricul-
tural students must be trained in power to analyze new economic situa-
tions as they arise, rather than suppliedwith rule-of-thumb procedures ‘
andsv‘anescent facts and data, has already been sufficiently emphasized.

But facts are needed as well as analyses. Teaching should
never unless absolutely necessary be in terms of hypothetical data. , The ‘
more realistic the data the better, the more nearly they describe the
environment from which the student has come and to which he will return.
The material for the problems of analyses that are set-up should be actual
facts and data. ‘ There are a150, other facts and data, sepecially. those
of a phySiCa1_ sort, which are sufficiently permanent in their’nature so
that the student should'be taught them thoroly. In this category are
the facts as to marketing channels for different farm products, the .
curves of output per unit of input for the different input factors in
the region where the students live, the usual conflicts between enter-
prises in the use of labor. The student whenhe leaves the cwnpps
should also take with him an understanding of the salient facts in the
present environment, even though some of these are highly impermanent.
Without this he will not be ready to put his hand to the pic?! in his
first undertaking. But even more important is it that he know about
the current sources of data of economic change and be taught how to

work them into new analyses of each year‘s operations, of each new
economic situation.

On the side of agricultural economics which relates to ques-.
tions of public policy, progress is restricted by a lack .of an adequate
understanding of the facts as to the economic organisation of agricul-~
ture, and the economic condition of farm people in general and in differ-
ent section, as to such public questions as those pertaining to taxation,
the credit needs of agriculture and their proper provision, "land utiliza-
tion and land tenure, the structure of land values, the relation of tar—
iffs and immigration to agriculture. It is highly important that the
facts as 'well as the analyses relating to these shall be developed and
taught to all agricultural college students. The required courses in
economics should include this sort of content above all else.

The foregoing analysis of the teaching needs of agricultural
economics furnishes the b‘asis’ for a discussion of the research which
is required for it. Strongest emphasis must be placed on research
which will develop the principles that constitute'the tools of analysis,
not the broad abstract prin‘tfiples of pure economics, but those Of the
applied science,‘ as defined by Marshall, of agricultural economics. It
is this most of all which is needed to give the courses in” agricultural
economics solid content, to-save many of them from the charge of being
thin or largely descriptive; Next in importance at present is research
which will provide an adequate understanding of the condition of agri—
culture and broad social queStion-s relevant thereto, Both of these
needs require looking at research on a wider basis than a. single state.
The research program confined within a. state, so 'far as teaching needs
are concerned, should be aimed at collecting data'of a. cross-section '
of the different systems of farming in the state, of marketing organ-
izations and marketing practices, of land utilization and tenure, of
credit need and facilities,’ etc. , so that these can ’be used as material
'for analysis in courses, and to acquaint students with the irportant
facts so that they will know their environment when they go out into
it. It should. also giVe them data of changing loCal conditions which
they can be taught to use in analysis. However, "if the local research
organized along these lines produces merely facts and data, does not
ermize these in relation‘to principles, the larger part of its Value
is missed. The courses using results of such research will be as thin
and a8 descriptive as those be generally condemned‘et: present.

As already suggested, it is desirable at times that much of
the re'sea'rch in some lines be built around the needs of specific courses
badly in need of being improved. Only in this Way can the teaching in
manylfields, notable among these marketing, be developed into something
substantial. _ ' I

"4:35!

".35
l: V

(1:) Relation to The extension service of a. land gram”
extension. institution needs to keep informed fully as

to conditions within the state and to follow

closely the changes in t‘: e economic environ-

ment. More important still, it needs to
understand. the fix of these, It is the job of the'researchs s t3 ff in
economics to collect and analyzea 1d interpret all this inf ornatim and
of the extension service to fit it into its current odu.cational programs.
The research program of the economics department should be sufficiently
elastic so that nhene1er en energ may situation arises, the extension
division can call upon the research st;=.f1 quickly to assemble the necessary
facts and analyze the . If the research ..taff is properly organized for
collecting current informati on, it mll usually be able to supply most of
t1 ~ information needed without sending men into the field.

The practice of working out a state program is becoming in“
creasingly general. Mr. Marquis of the Office of Information of the
U. S. Bureau of Agricultural Erronor'ics reports that in 42 states the ex-
tension service made spacial a;1‘e:-._emen ts for disseminating the national
outlook report. Sich prcg ’I’Uns .hould have a short—time or “outlook"
aspect, and also a long— time aspect. Once such a program is adopted,
it is the business of the extension service to present it to the people
of the state, and of the research divisiOn to carry on the research fun-
damental to it. This means a contimling program of research, an import-
ant part of which is the analysis and. interpretation of current develop-
ments, The researcl. staff should take a major part in the formulating
of "both the longv-ti: anrf short-time features of the program. It is
their records and their in terpretation of them which will furnish the
basis for the outlook stateirents.

Extension work is becoming much more frankly educational than
it was. More use is being made of schools. One of the important func-
tions of such schools should be to present the results of the most recent
research projects in the same or other states.

Extension work in the past has been to largely a matter of
talking - of getting groups of farm people together in school houses
:nd town halls and talking to them. Extension men have often been
selected because of' their ability as talkers. It is now taking on
‘ more and more the nepect of programs, "drives" and demonstrations.

A pro gram or "drive" is something that should be planned carefully in
advmce, frequently requiring a certain amount of researches a. founder-
tion for it.

fiEl-fiflfl-QEE Regulatory bodies at the start seldom
regfl atory ‘ realize the importance of a real understanding
activities of their problem such as can be had only after

Careful research in the foundations of it. .
ibis has been illustrated many times in the experience of- the United States
Department of Agriculture - of late, particularly by the Packer and Stock.-
yards and the Grain Futures administrations. The men selected for these

1.5“.» .

”426

administrative roles are selected principally, as should be, because
of their administrative ability. But it is possible for one to be
a good administrator and yet appreciate the function of research and
have some understanding of its methods. Of course all such adminis-
trate rs would deny absolutely the charge that they do not appreciate
the importance of research to their work; one has to look to the re-
sults of their administration to discover the true situation. Often
a. research department is set-cup; but for some reason or other, either
because of the men selected for it, or of the tasks assigned them, it
too often fails to produce any effect on administrative policies.

(6.) Relation to fgt- The job of fact-gathering is
_ _ gathering activities. most expeditiously and economically

 

done when it is standardized and re-

duced to routine. Changes are expen-

sive and bothersome. New departures
are alWays experimental. The research man does not recognize all the
difficulties in the way of collecting certain information. He thinks
of what he would like to have, and is inclined to assume that. the rest
is merely a matter of perfecting the technique for it. Not analyzing
the data himself, the fact-gatherer is not spurred on by the same ob.
Jectives as the research worker. His role is that of assembling
material for others to exploit.

W , It is one thing to recognize the

W ' - relationships between fields of work;

(a) W it is another to secure the coordinated
321-933 effort of those engaged in the related
-Wo fields. It may arise between them in

. a spirit of pure cooperation; it may be
forced upon them, with doubtful results, by an autocratic administration;
or it may be accepted by the parties concerned merely as a matter of ex-
pediency. . securing such coordinated effort quietly and painlessly is
one of the achievements of a good administrator. Proper organizational
arrangements will assist greatly to this end. This section will discuss
some of these arrangements.

Some institutions favor formally stated joint projects between
departments; others having one department in charge of them, but the others
helping out when their help is needed. Still another arrangement is to
have separate projectsl statements prepared, but so coordinated as to give
results‘in" the two related phases of the subject. Thus, for example, the
animal husbandry department might outline a feeding experiment looking to
providing an input-output series, and the economics department 9. project .
collecting-and analyzing the economic data needed for combination (with the
input-output data. No doubt each of these arrangements has its advantage

 

*Professor 0. L. Holmes contributed portions of this discussion.

and special suitabilities. Where one main objective is involved,"and
lies Clearly within the field of one research division, the second. ar-
rangement is probably the best. If there can be Said to be two definite
parts to the. project, even though they be closely related, the first or
third arrangements will work the better.

‘ In some institutions the economics departments find getting
together with the natural science. departments on a joint project in
which the contributions of each are properly recognized, a delicate
problem. Gheyork in economics is new. In many eases it has taken
over certain more. or less disputed territory, which previously has been
considered the domain of various natural science departments. In not
all Cases are the administrative officers of the institution fully ap-
preciative of the legitimate scopesnd nature of the economics work.
Perhaps if an inventory were taken of the situations within the various
experiment stations, there would be found what an unbiases outside party
would consider an entirelydnadequatej view of agricultural economics
work and the extent, to which it needs .to be handled by persons specially
trained for it. . But gratifying progress indeed has been made in this

reSpec‘t in the last ten years. , g

,_ ‘ _ ferhaps.‘the mostfiifficult problem of all is to get natural
science departments to modifymheir methods so as to get the results in
form, so that economistscanruse them. {If the economics department fails

in this effort, Iit become-ea responsibility of the director to take a
hand in it. H H I H ,

In the case of newly developed ”lines ,of research, some import-
ant questions of organization arise. In "the case of research in the
economics of home economics, it is to be doubted if two entirely differ-
ent departments of economics, one for the farm side of the combination
and the other for. the family or household side, make a satisfactory or-
ganization. A better arrangement is one‘ldepa'rtment with certain members
of it delegated to work on the household ‘andfamily problems, these
persons having received training in both economics and home economics.
Another arrangement is to have some trained economists on the staff of
the division of home economics who work in close Cooperation with the
division of agricultural economies. ' " '

It is equally to be doubted whether a separate department
should be set up; to provide. researchin the economic. or reg-"try.
Projects of thisnature in land grant experiment stations should be
worked out by conjunction of~the two departments. The men selected
for it at the start-should-probably betrained economists who will have
the help of teclmicalforesters at every turn, and who will gradually
acquire a'workinghiowledge of technical forestry. As soon as possible,
however, a considerable number-bf graduate foresters of special ability
should take up graduate work in-economics .in general departments of
economics, supplemented by certain courses in agricultural economics in

order to get the points of view of applied economics. '

428

(b) Research 3 The first administrative problem here

and related is that of a proper distribution of emphasis
functions. among the three major functions. In many

institutions extension work has been pushed

in advance of the research effort as a basis

for it. Extension men have been appointed
in many instances before any research has been done. The writer recalls 9.
visit a. few years ago from two rather eager young men who had been appointed
as extension men in agricultural marketing in a state where there had been
at that time not a single project in marketing research started. They
were seeking for suggestions and for material which they might use to make
their extension work effective. In another state at the present time
there is approximately twice as much staff devoted to extension work in
cooperative marketing as to teaching and reaching combined in the same
field. ,

(l) Reseggch and The important problem here is whether
teachig. ' teaching and research should be in separate

organizations entirely,carried on by men

doing research only, or whether the same

department head should have charge of both
both and both are done largely by the sans staff. 1310 letter Wt
is the more usual in the field of economies; but not necessarily the better.
Some very capable teachers are poor research sorkers; and some good research
men are poor teachers. Specialisation that will take advantage of these
special abilities and limitations is highly commendable. But this amount
of specialization is possible'without having separate administrations. It
is always possible to have some of the staff working altogether at teaching
or research.

However, Profesaor Holmes, writing on this subject, doubts whether
any one should specialize entirely in research or teaching. "A teacher will
have his subject—matter vitalized and his contacts with students more stimu-
lating' and effective if he is given time and funds for the prosecution of
research. A limited amount of teaching, on the other hand, tends to de-
velop a saner point of view on research and a more vigoroug prosecution of
the project. It is impossible, of course, to give each research men a
teaching assignment. But it would seem the part of good organization
wherever possible to have each of the teaching staff’directly responsible
for some research project and to expect from him a reasonable showing both
as to the quantity and quality from his research efforts." It is highly
pertinent question nhethervone who is poor at research is really a high-
class teacher. He is likely to be a popular rather than an effective
teacher. 0r his abilities may be wholly in elemtary courses. Unc- '
doubtedly most teachers, especially-of advanced courses, benefit greatly
from research- experience. ‘ Some of their best teaching grows out of it.
More important still, if the teaching is to' be vital, the teaching staff
needs the close contact with problem and with conditions out in the
fields which research gives. Nothing is more fatal than to have a.
teaching staff getting all ideas from books.

429

Likewise some research workers are improved even as research
workers by the experience of presenting their material to students.
Research is likely to be narrowing. The workers confine their activi-
ties more and more to special subjects, and presently find their use-
fulness in handling these various subjects greatly reduced by their loss
of vision of the larger field. Teaching a few courses which keep them
in touch with the whole field is an excellent corrective for this. Re-
search workers freely testify that many of their best research ideas have
some to them in the classroom. '

The foregoing is especially pertinent in the field of economics.
As will appear later, careful deductive analysis of research problems in
ecOnomics is highly necessary at all stages, especially at the beginning
and end. Working closely with concrete data year after year has a tend-
ency to numb one's analytical facilities. This accounts for the tendency
of the research work of research specialists in ecouomics to become in—
creasingly fruitless in later life. That this tendency is a real one,
such research workers themselves have abundantly testified. They are
always looking for a chance to do a little teaching; to get away from their
work for a year of study. '

The largest objection to having economics research and teach—
ing ine one department is the difficulty of getting an administrative
head who is sufficiently inclined toward both. But this difficulty
is easily avoided if the danger of it is sensed in the beginning. If
historical reasons have put a person in charge of a department who is
poorly qualified to direct research. In such a case, the best pro-
cedure may be to have a separate head for the research.

Even the foregoing danger is slight in comparison with that
of having a teaching and a research department which do not work to-
gether harmoniously. In the early stages of a. department of agricul-
tural economics, it is especially important that the research be planned
so that it will provide teaching material. The chief administrative
officer of an institution can bring the research and teaching departments
together and keep them there; but it may take constant vigilance.

,(2) Resegch and It is probably more commn to have the
extension, extension work in economics under a separate

administration than it is the research work.

Most extension directors feel that they must

have a staff of men of their cm at all times
available that they can shift from point to point in the state at will.
The also like to have a staff entirely their own for the sake of the
better esprit des corps that is possible and so that they Can infuse it
with one set of ideas and points of view. If the members of the exten-
sion staff are also members of subject-matter departments, they are likely
to have strong ideas that they get from their associations in these depart-
ments, and the result may be a less harmonious program. Of course this
latter would not seem a. disadvantage to e. broad-gauged extension director.

 

430

The strongest case for an entirely separate extension staff
is usually made on the score of specialization; that in the first place
a man with special types of ability and personality is needed, and in
the second place, that years of experience are needed to make a really
good extension worker. It is commonly said that no one should be em-
ployed as an extension specialist who has not been a county agent for
several yearS.

There is also a strong disposition for some extension divisions
to be sufficient unto themselves in economic matters; for the director
in particular to assume that he is fully conpetent to follow what is
going on in the economic sphere and make the proper interpretations of
current developments. An economics department which confines its re—
search entirely to pure research or to social problems leaves the exten-
sion service in the position there it has little else to do but attempt
to follow and analyze current developments itself. As already pointed
out, inadequate support may make it advisable to conduct the economic
research along these lines. But any land grant institution which
forces its economics department to follow such a pol-icy, is extremely
unwise. No doubt in many cases the extension division is partly re-
aponsible for such a policy. If it is recognized that its proper flinc-
tion was limited to pure extension, that a properly sustained research
staff in economics sould do a. much more thorough and scientific job of
collecting the necesaary information, and certainly make a sounder in-
terpretation of it, it would insist that the administration give proper
support to the research staff in economics. That such is the ruling
opinion among extension leaders is demonstrated by the language of their
recent report.‘'

One of the reasons advanced against such a policy by some ex-
tension directors is that the research staff is not closely enough in
touch with conditions in the state to make the proper interpretations.

The answers to this are three: First, a properly supported staff would
be able to keep in touch with conditions in the state. Second, to
obtain this information the research staff would not need to be travelling
constantly as are the extension men. They would collect it by more
scientific methods than sketch interviews with people encountered in

the field. Economic information cannot be observed growing in the field.
It cannot be obtained from the opinions and attitudes of a few individual
farmers or bankers. Even the county agent is often a very unsatisfactory
observor of economic conditions. He meets farmers every day with re-
actions as far apart as the poles. The choice which he makes between
these reactions may easily be highly personal in its nature. County
agents in adjoining counties often give greatly varying reports of
economic conditions. What is wanted instead is records of conditions,
and these need to be collected under scientific supervision. The place

 

* See page 1.

4:31

to get many of the records is not at the farm, but in the market.

Others of them can be obtained by an adequate reporting service. This
is not to-bexi'nterpreted as saying that the research staff never needs
to get out into therural districts; but most of its travelling will

be in connection with the surveys which should be constantly in progress.
One of the things that'needs to be surveyed is attitudes. What farmers
think, and what they know about economic matters is basis to any effect—
ive extension program. The first maxim of pedagogr is to proceed from
the existing or known to the related new or unknown, Third, what
strictly current information the research staff needs as to famers'
attitudes and the like, it should be able to obtain from the extension
workers if there is proper contact between them and the research staff.

Another common arrangement for extension work is to have it
done by people who are also members of the staff of departments of
economics, but who give all or part of their time to extension activi—
ties. For the subject-matter which they extend, they are responsible
to the head of the department of economics, for their arrangemnts in
the field, to the extension director. This means that the extension
program for the state is'worked upby the regular staff of the land-
grant institution, primarily by the heads of the departments, in council
with each other and the special administrative staff of the extension
department.

The advantages of this plan are as follows:

1. The extension program is likely to be better founded and conceived
because the strongest men in thevinstitution, who are at the same time
in close contact with the research work, participate in working it out.
Under the first system described, the extension director seldom has
enougo help-from the economics department. Too often he does not feel
that he needs it.

2. Incidently thereto, the reSearch and extension program are brought
into line with each other. It is possible to accomplish this under

the first plan or organiZation, but it is seldom achieved.

3. The whole institutionis behind the program because all are fully
informed concerning it, and all have helped formulate it. Under the
first arrangement, the teaching and research staff often know very
little about What the‘ext‘en'sion staff is doing.

4. The teaching and research men, in particular the heads of the
departments, are kept in' constant touch' with conditions and people

in the state. Even if the do not get out in the state on extension
work very mich themselves',"they have their offices next to those who
do and are in frequent discussion with them. This makes the teach-o
ing and research more vital and more responsive to the needs of agri—
culture. At the same time, the extension men are benefitted greatly
by frequent contact with the teaching and research staff.

432

To make this second plan more effective in this latter re-
spect, 'it is-bnly ~ne'CeSSary to mix extension, teaching and research a
little and' send members ’of the teaching or research staff out on ex-
tension as'signrnen'ts occasionally. It is highly desirable that the
men sent out on important assignments should be the strongest and
best informed men in' the institution. Economics is so easy a sub-
ject to go astray with, and economic questions are so" constantly in
the foreground, that this becomes a highly impertant consideration.

5. This-leads to perhaps the most important advantage of all, that
under the second plan the men who come to stand in the eye of the
people of the state as representing the different fields of agricul-
tural endeavor, are the strongest men in the institutions, those who
either engage in research themselves as part of their work or are in
close contact-with it.

In contravention of the foregoing, it may be stated that
the extension department may have specialists in economics and other
fields who are also strong men. Conceivably, yes; rarely so, in
practice, and for rather obvious reasons. First, there is not room
in most land—grant institutions for three strong men, one to head.
teaching, another research; and another extension. If one strong
man is available for the three combined, the institution is fortunate,
unless the institution' is one of the larger ones. There are reasons
why the teaching and research men are likely to be the stronger. Ex-
tension work in the past has not attracted its proportion of really
able men. Most of the men of ability in educational work today have
been attracted to it because of the fascination which scholarship has
for them. Extension work has thus far discovered no incentive which
is equal to that of scholarship for the usual person of ability. Those
Special fascinations which it does possess are probably better rewarded
in ,business, politics and jounialism.

Second, the experience of being an extension specialist does
not develop strong men, in the sense in which this term is used in the
scientific world. A considerable ammmt of such emerience is recoog—
nized as very valuable. But in the end it leads away from that careful
study and close analysis of problems which alone makes for intellectual
growth. There is a tendency for many extension specialists as the years
go by, to pride themselves more and. more on being “practical", and to
think more and more frowsily, until they are scarcely distinguishable
intellectually from the common herd. This is"especially easy in the
field of economics, where loose thinking is so easy and so common.

That which will save them from this fate'is constant close contact with
research workers, supplemented if possible by-periods of study and actual
teaching and research work.

As already pointed out, extension work these days is more a
matter of working out programs and organizing them and sitting in coun—
cil. The abilities and training required for this are much more nearly

433

like that of teachers and research workers than that required forthe
old-fashioned farmers' institute'type of extension specialist.

It will be freely granted that men who are to (do extension
work regularly should have had county agent experience if possible. .'
This is just as possible under the second plan as under the former.
There are hundreds of experienced county agents each year looking
for a chance to get into graduate work.

The importance of the foregoing considerations is being
increasingly recognized. One way of trying to meet them at institu-
tions where the experiment station and extension staffs are separate
is to arrange regular weekly or monthly conferences between the two
staffs. More and more it is coming to be the practice at such in-
stitutions to assign station men to an oceasional appointment in
order not only that they may keep in- contact with the farmers' inter—
est in the work they are doing, but that they may, in some cases at
least, lead in a demonstration of the use of new material. At one
institution, plans are being worked out by which the extension workers
will occasionally be assigned for a certain period of uninterrupted
time to actual investigations]. work on the progects of the experiment
station staff. _. . . ,

(3) Research and _ ' Should research that relates to regu-
re ate 1:: tory functions be delegated to separate
activities. research agencies or administration?_ No

general answer can be made. Analyses of

‘ - the fundamental sort should usually be
turned over to separate scientific agencies, such as the research
staff of: the department of economics of a land—grant station. That
which relates closely to current administration needs in‘ nest cases
to es done by research men employed by the regulatory staff. A ‘
very desirable arrangement is to have a capable research men as one
of the regulatory body. He will know what it is best for his staff
to undertake and what is best to hendlein cooperation withseparate
research agencies. 7 -

It is extremely easy to be unfair to regulatory bodies
when they refuse to Open their records to outside agencies, even to
departments in agricultural experiment stations. If they knew the
limitations of our departments of economics as well as we ourselves
know them, they Would be ev-en more unwilling to offer the use of
their records. There should first be a high order of assurance»
.that the proposes is well conceived, that the research men are com-
petent to execute it, and that the results will be carefully re-
viewed before publication.

If regulatory bodies would only keep adequate records of
their own activities, research in their problems would be greatly

434

aided. Frequently it is only when some competent outside agency
begins work on their problems that they discover the weakness in
their records. This is often true even when there is a "research
department“ in the organization. By its records, you shall knowhow

fully a regulatory body appreciates the research aspects of its prob-—
lems.

(4) w £111, Overcoming the difficulties in this
Fact—gathering. field is a matter of organization. Arrange-

ments can be worked out by viiich the research

workers and the fact-gatherers are brought to-

gether and made to understand each others ob-
jectives and problems. The fact—gatherer should be made to understand
that he is an important agency of economic research, and should be given
much more credit and recognition in publications than is ordinarily the
case.

The difficulty that the data are not gathered with broad
enough objectives, can be met by holding conferences of interested
groups. This might also win larger support for appropriations for
such mrk.

Developments in the field of gathering current data are only
at their beginning. Each new step forward in research reveals new
data that should have been collected since the beginning. That the
data have not been collected is often more the fault of the research
workers than of the fact-gatherers. The research Workers have not
known what they want. They do not yet know what they want or have
not yet worked out satisfactory measures.

But depar,tments'_of agricultural economics lmow of a great
deal of data which in most states are not now being collected, or are
not collected insufficient detail. Data of local prices, of land
sales, of rents,‘ of taxe's,‘ seasonal marketings, movements of popula—
tion, of unemployment,' disease infestations, ammts of labor hired and
feed and fertilizer bought, sales of machinery and building material,
bank loansm interest rates, rate o‘f‘payments on mortgages, land im—
provements, yields,, acreages of minor crops in trucking areas, etc.
No doubt the biggest problem is making arrangements for collecting
much of this information. The first line of attack seems to be to
expand the cooperative crop reporting services of the states by
securing more liberal swport for them.

(0) Research £1.11 ' There is also the question of the
one department interrelations of the various problems

within the field of economics. Where,

forexample, does a marketing project

leave off and a production project begin?
What shall be the division of effort between the man studying rural
finance and the one studying farm management with reference to such

435

problems as farm investment, mortgage policy, and the like? These
and other problems of a similar nature confront the man who finds
himself.- responsible for developing and carrying out a well—rounded
research program in agricultural economics designed to yield a mass
of product for teaching, for extension, for the guidance of business
executives and farmers who are looking to us for aid, and for the
general. enrichment of the bedy of information in this field. It is
not doubt easier for two or more persons in one department to cosper-
ate than for separate departments to cooperate.- It is easy to pub-
lish the results of each joint effort in such a way that credit is
received where credit is due. It is also easier for administrative
authority to exert itself unobtrusively and lay out the work in such
a Way as to secure coordination. Nevertheless, friction is not al-
together absent even within departments.

4. Accrediting ' ' ' The first question to be considered

results. here is what is authorship of a research
'. ' ' study? It surely includes both the analysis
of data a‘nd'th‘e preparation of the report of
' it. It is rather hard to separate these two.
A man may do a considerable amount of- analysing and yet write not a".
paragraph of the document in which the analysis is presented. On the
other hand, it is hard to see how the man who actually writes up the
material can avoid participating in the analysis itself. (Jon-inesiti'on~
is an essentialstep -4 the final one - in analysis. '

The mere process of fast-gathering, freely participated in
by junior members of research staffs, by field men and sometimes by
mere routine workers, can hardly be construed as authorship. That
their work should be acknowledged at least by footnote reference is
obvious. But often there must be a close relationship between the
gathering of the original data and the analysis and presentation of
it. Much of the s'i'giificance of the raw material of certain studies
must be conveyed through the impressions received by the field men or
other original investigator. Where such is the case, he should be
credited forexactly this sort of a Contribution.

A more difficult Question is whether the chief or assistant -
chief of a research section, fer example, shell put his name upon a.
publication rarely because he has conceived the problem, has edited
the manuscript with more or less minuteness, and given general direc-
tion to the prosec‘tion of the study and the preparation of the manu-
script? It seems to the writer that this should ordinarily be answered
in the negative. ‘ on the whole, it is profloably, the part of wise ad-
ministration to be as libenal 'as pOSSible in the granting of credit
to those vino are taking care of the details of the work in responsible
fashion. ' , . ,

In this connection also arises the prObleIn of joint author»-
ship between members of the some staff. ‘ It seems the part of wisdom

‘33

436

F

to keep authorship as separate as far as possible, particularly for the
sake of the young and inexperienced research ‘wrorker whose reputation is
yet to be made. The appearance of his name along with that of a man
who has already published may mean but little to him in the way of
recognition. ‘ The credit in the mind of the reader will almost in-
variably‘go to the senior author -- the man as has already demonstrated
his ability to study and to write. It is frequently possible to divide
the results of a comprehensive project on which several men have worhd
into separate publications, each bearing the name of the tone of the
worker who has'contributed most to it. -

Where separate departmentxfiork on'the same project on .a
joint basis, only some form of a joing authorship will meet the situa—
tion, unless separate publication seems possible.

Relationships Besides the relationships to other
to resegch in departments inthe same institution are
other institutions . those to outside institutions, to other

land grant stations, to. state departments
. of agriculture in the same state, to the

U. S. Bureau of Agricultural Economics, to research institutes, etc.
First of all is the possibility of a vast amount of duplication.
Second is the possibility. that one agency will attempt to do something
that another is better placed or organized to do. Third, that import-
ant gaps will be left in the research on certain large problems, or the
converse of this, that agencies well situated to handle certain phases
of problems will- not be called upon to do it. Fourth, that there will
not be team work on problems requiring it or benefitting greatly from
it. - .

Let us first illustrate this in the case of the land grant
stations. If a large group of experiment stationsin the Cotton Belt
all undertook to analyze the broad national problems of orderly market-
ing of cotton at the same time independently, there would be a great
deal of duplication. They would all be seeking about the same sources
of data and going through about the same analytical steps. If, however,
One of them best situated for it'worked on the national phases of it,
and the separate cotton states could be interested in collecting and
analyzing data relating to important local aSpects, then there would
be no duplication, and in addition a preper division of‘la‘bor with
probably no gaps in the program. If the several. stations were in
addition to plan their local attack so that the data Could be combined.
then there would be team work in addition. Stations with central.
markets close‘at hand can carry on much of the research in these markets
more economically and understandingly than those at a distance. However,
for them to be responsible for all such research is altogether too much.
Moreover, in somecasesit is highly advisable for those working on
marketing problems in the producing territory to follow the product in-
to the central markets to see how it is handled there.

437

Duplication. is less likely in economic research than in most
other fielder Local differences‘are often one of the important as-
pects of economic research. I A study in one state often" needs to be
repeated in several others in order to bring out all the aspects of it.
Thus, for example, the Organization and management of country grain
elevators needs to be studied separately for regions handling mostly
wheat, mostly corn and oats, or different combinations of coarse and
bread grains; also for regions where handling supplies is an important
adjnct; and for regime where buying and selling, hedging and storing
practices are different. Problems of intensity in the use of factors
of production need to be repeated in all important areas and farming
systems in which a given crop or kind of livestock is raised. Out of
these numerous studies will finally emerge larger generalizations that
will be of use to all ‘ ,

On certain new and difficult problems, such as some of those
in price analysis, it may be excellent general research policy to have
several agencies working at the sometime. Cdmpetition has its value
in research as in other fields '- if it is not overdone. One group of
workers is likely to have a particular approach so clearly in mind that
it overlooks the right one.

The relationships in research with the U. S. Bureau of Agri- -
cultural Economics are too generally recognized to need much discussion.
The most important aspects of it relate to division of labor, to having
the Bureau handle the. phases of it that it is best qualified for because
of access to important data,- or facilities for collecting it. 'Any pro-
.ject requiring data over a- large part of the United States needs to be
carried on by or in close alliance with the Bureau. Likewise projects
requiring foreign data. , The other; important a8pect of this relationship
is that the Bureau is in the best position to sense many of the larger
research needs and henc'eto' assume leadership in getting projects re-
lating to them under way; The. Bureau is also ”able to help many of
the newer and smaller departments with research-methods.

Other aspects of this problem have been sufficiently dis-
cussed under an earlier head. .

Rglgtionswith' ‘ . ' ‘ I .Ooeperationwith such agencies as

W ' - ‘ . . institutes" of research is largely'a matter

AME-fl. . of division of labor in the field of re-
‘ ' " -" ‘ search, already discussed under another
head. ' Some of the broad problems upon
which research institutes work may lend themselves to cooperation
with state institutions which are in close contact with certain
aSpects ofthemv

Some of the most difficultflp'ro'blems 'of joint-effort in. many
states center around state departments of agriculture of markets. The

438

statutes creating such‘departments sometimes limit their sphere of
activities so as to- exclude research, but usually they do not. Even
if they do, there is nowhere any definition‘of research which is so
clean—but that it jean be uSed for settling questions of boundaries.
State departments need research for their regulatory duties even if
they do no" research for its own ends. All, that'has been said earlier
about the relation between research and regulation applies in this
connection. In view of this,‘ it is highly desirable that cooperation
be arranged whenever an urgent problem presents itself in which both
agencies are interested. In other cases, the simplest solution of
the problem is always to take up with the other body any new venture
in research to make sure that there are no overlapping plans.

W’ The preceding sections have discussed

W3 the relationships between the research act-

1- M" 1 ivities of various agencies and the needs
ham and circumstances of joint effort in re-
Ml . . . search. i‘his section will undertake to

show how this joint effort can be best re- '

alized. ’ The conclusions are based on the
experience of a considerable number of workers who furnished reports
and suggestions. ~ -

In general, there are four bases upon which joint effort
may be arranged: (a) coordinate departments or institutions cooperating
on a more or less equal basis, (b) one department or. institution in ’
charge of a given project, but calling upon others forrelated phases
of it, (c) an outside agency, such as the U. S. Bureau of Agricultural
Economics, acting as leader in the project. (d) a superior administra~
tive authority exercising a sufficiently detailed and rigorous control
over the whole project so that the various phases of it, althou$1
handled by different agencies, are held in line with each other. Which
of these four works most satisfactorily depends upon the personnel, the
project, and the agencies. Cooperation is the loos'est of these arrange-
ments. In general, purecooperation will work when the common end to
be obtained by joint action weighs larger in the interests of the workers
than the individual ends to be gained by competition. one may recognize
that certain important projects require joint action; but there is always
the possibility of finding some project less important socially or scien—.
tifically which may yield larger returns to the individual performers,
personally or professionally. In its pure .fo rm, cooperation assumes
agencies on the same-plane working together on a. basis of equality and
sharing in the results in proportion to. their contribution. . In its
pure form, however, it .seldom exists. What passes for it instead is w
having an agency taking the lead in the project and the others helping
out whenever asked to do so, credit being given in proportion to the

 

* By Professor C. L. Holmes, Iowa State College.

439

contribution. The next step beyond this is when an organized attack
is set up in which one agency or person is definitely in charge of

the project, and the job is definitely divided into tasks which are
assigned to the "cooperating" agencies. It is obvious that as soon

as one person or agency steps into the role of director, the basis of
pure cooperation is lost — the cooperators begin to look upon thepro~ ..
ject as somebody else's in which they are helping.

The specific problem before us is the exact procedure which
may best be followed in formlatinga project in which two or more
cooperating institutions are eXpected to participate. Hc'm go about
it to have each of thecooperating parties make its best contribution
to a common enterpise? How avoid having the project made one-sided
or inadequate by one or another of the parties failing to assert its
due amount of weight in the councils?

One of the first objectives in the getting of a good project
is leadership. The first question asked should therefore be: "Where
is the best leadership for this enterprise, either within our own staffs
or elsewhere? How can such leadership be brought into service, not
only in the actual carrying out of the project but in its formilation?"
Several of the correspondents stressed this point in their replies.

One of ,the first things implied in the development of a
project of research on a cooPerative' basis is that of conferences.
Conferences, if rightly conceived and carried out, with the background
of prOper preparation, are absolutely essential to the development of
successful}. coOperative programs. On the other hand, much >w1uab1e
time may be wasted and poorly conceived plans may result from confer—
ences not preperly prepared for and carried through. ‘To be successful,
conferences must be limited in scope and have their objectives clearly
defined in advance. It is also important that each participant know
definitely in advance the specific things to be considered and devote
some real thingain to the problem before he enters the conference. An
exchange of ideas almost invariably results in a. clarification and a
more comprehensive conception of the given problem than could result
from the work of a single individual. 0n the other hand, the expression
of vague notions and ideas is a most effective means of Wasting time.

Many of the correspondents expressed themselves in favor of
Conferences: To quote from a Middle lestern department head:

"Policies and programs should be worked out
by conferences attended by representatives of each
state and of the Federal Bureau. If necessary for
such conferences, the Federal Bureau should provide
travel funds and bring representatives of states

together at convenient points for these conferences.
This appears to be necessary since many states either
restrict or prohibit out—o,fg-state travel by their
station workers." H ’ '

a In states in which little work in this field has been done,
it has been necessary that the Department of Agriculture assume leader-
ship and furnish the initiative. Many times this was necessary be—
cause the Department of Agriculture had a program of its own, the ad-
euqate carrying out of which involved work in states inwhich but little
or no work had been done and in which, for the am. being, adequate,
leadership was not available. There is no question but that great
good has come from such a procedure, but-occasionally in the corres-
pondence shots of criticism is detected. Probably the point in. the
following is well taken: - -

"Ihe only criticsm that I would make of such
cooperation is that when someone in Washington gets
an idea of a. certain kind ofwork, this work is , -
started all through the country because it is sub-

'- sidized and receives a. prestige based on finances
'rather'than on merit. ‘ Itithen comit-s a. lot of states
to a lot of Work which’may not be the most important
work in that state at that time. Twice I have‘seen the
cost accounting fever run through the country. ‘ If I
recall correctly, practically all of themany cost .
routes in the country were in part at one time supported
by the Department of Agriculture.-——-— Only one or two
states were doing this kind 'of work without a subsidy. "‘-—---

.‘Mnoth’er objection to cooPeration is the fact that
plans in Washington change so frequently that work is no
more started before some other’plen develops.3‘

‘ It should be added that with the expanding needs of the out-
look work of the Bureau ofpngricultural Economics, and the increase of
state funds through the passage of the Purnell Act, the federal Bureau
haswisely withdrawn most of its financialsupport to state projects,
particularly to those localized wholly within a state. But there is ~
still possible the same participation in planning of research.

During the last ten years, which period has witnessed the
most of the growth in c00perative agricultural economics research be-
tween the Federal Department and the several states, the Department ,
leaders sought to systematize their cooperatiize relations in'sev-eral

 

I"Du E. W. Allen's cement on.this statement was that "It is only
fair to point out that the activity of the Department Was to a
considerable extent in response to pressure from the states and
"Congress", that "the states had the cost-accounting fever too."

441

..

parts of the country by the establishment of the so~called "research
councilflx" The most outstanding example of. this type of cooperative:
procedure is. the New mgland Research Oouncil.--I This :Gmmcil is com-
posed of representatives of the agricultural economics work in the '13
Various New England states and maintains a;- central office in Boston; '
which is in charge of a. representative of the Federal Bureau of Agri-
cultural Economics, who is designated .as its secretary. According to
a letter received from the present secretary. of this Council, the pro-'-
cedure in outlining work is as follows: , . ' .. .

"'Ene policies and program of the Research Council
Council are formulated by the following steps: first,
they are discussed with the research specialists in the
various colleges and by representatives of the Bureau
of Agricultural Economics. I gasped,- a program is dram
up in a tentative form by the secretary of. the Council.
Third, this tentative pregram is again discussed with I
the various interested partiee and changes are made from
time to time. .- Fourth, this meg-am is presented at the
annual. meeting of the Council which attended by most ofI
the research Imen in the Economics Departmnts of the
colleges in New England. " I’ II 6.3. ..

Both connendatory and; somewhat critical cements have been
received as to the success with which this Council has realised its .
obJectives. Qne criticism was to the. effect that the work suggested
had been of Is. superficial rather than fundamental order, that the
conferences had been too brief and had gone too little into the de—
tails of the preb1omsI which it was proposed to stuiy, and that the
publications which have. resulted from the series of projects which .
have been carried out ass. result of the work of the Council lack
many features, of good research pullications, and include certain
recomendations and conclusions xio't warranted by the limited facts
obtained in the investigations. .- -: 4 -..; .
In contrast to: the foregoing, another leader from the New.
England states speaks in highest terms of the effectiveness of the
New England RIeISBaIrch CoxxeciL. . He. says: 1 , ,. .-.
"plans and possibilities of cooperative action and. the
possibilities -of coordination Ief activities of different
states ,are dismissed in the meetings of the GoI'ImIcil. It
does not foster projects in the senso that it attempts to
superimose regionwide pregects on the members of the ICIIO‘tm-
ciIII. . It is not madame, to a highly centralized co- , '
operative but rather is comparable to e. cooperative organi-
zation of Ithe federated type with control largely if not '
entirely in ”the hands of the constituent m‘eziber's' ahd with;
the Research Council working along those line which the'
members agree in conference are lines to be followed. By

442 5

“general agreement, however, if the executive secretary of

the Council knows that three or four states are particularly
interested. in a definite problem - we will say milk marketing
orgmiization or taxation - and if he finds that it would be
agreeable to the workers on these projects or the administrap'
tors in charge that a. conference be held' he" is able to bring
such a Conference about usually at the suggestion of one or
more of the interested parties. ‘ ‘ '

”The executiVe sedretary, es a competent representative
of the Department of Agriculture, is also" able to act as a
liason officer betweenvthe Department and each of the states
or all of the states interested. in a particular project.

"It is not the policy of the Council to foster research
projects or to try to enlist the states in particular project.
Where the representatives of the states in Council matings
are agreed that a. particular project is likely to be developed
in each of the states and that the project would profit by an

inter—state division of labor - as was the case in the apple
studies and in the dairy studies ~ then the executive secretary
is instructed to give all possible assistance to help coordinate
the work of the different states along the general lines 'of the
policy laid down by the Council.

"The reasons for nuccess of the Council, I would say,
are as follows:

"1. 'A definite spirit of cooperation exists among the
members rho recognize the advantages of‘ cooperation and coord-
ination of effort.

”2.176 have been very fortunate in having executive
secretaries who recognize the necessity of each state's main-
taining its administrative autonomy but at the same time who
recogxize the value 'of such an organization as the Council.

"3. The Council places at the disposal of all its
members the facilities of the Department of Agriculturenvhich
it has been difficult for us to comand particularly because

‘ of lack of close, first—hand contact with the Department."

For a few years, an organization .similar to the New England
Research Council was set up and maintained in the Middle West with an
office at Chicago. Later another was sat up on the Pacific coast.
Both have been abandoned. No doubt an important reason for their
failure is that the states in the West are large enough to be able
to operate by themselves more effectively, so that they feel less
urge to cooperate. For several years,‘ the Department has maintained
the Ibod MarketingLResearch Council in New York City. The objectives
of this organization are in general similar to those already outlined,
but it restricts itsinterests to a narrowerra'nge of problems and to
a narrower area.“ " .

443

There is little doubt that the federal Department may be the
most promising agency for leadership in getting the various state in-
terests tog-ether on a research program covering fields of work which
obviously'are regional rather than merely state-wide in their scope
and ramifications." One state man writes of this leadership as par-
ticularly evident in a type of study in which grape growers, been
growers, apple growers, vegetable producers and poultrymsn are in
competition with each other in several adjoining states. Production,
marketing and other studies need to be made simultaneously in the
different competing areas so that farmers in the different areas may
adjust themselves more quickly to competitive conditions. In some
cases, this means that a. certain area will discontinue producing a
certain'product. If~it is to be driven out, it is” to the distinct '
advantage of farmers to know that at as early 9. date as possible.

Another point is that certain problems cannot be studied in
their entir'ety‘wi-thin'-the:'~border_s' of agiven' state. For this reason,
studies limited to asingle state are often misleading and fuitle.

Another state man has this to " say' on the point:

"While I Was in the Department of Agriculture and had an
opportunity of observing the farm management work in different
states, I was deeply impressed with‘the‘possibilities of at-
tacking problems On a regional'basis on the assumption that a
great many economic problems have no relationship to state lines,
but are confined to productive areas, and that much geater pro-
gross could be made as far as both research results and extension
activitiesare‘concerned if the problems could be attacked on a
regional basis." _ , , , , _'

The Department is to be warmly commended for its earnest and
persistent effortsto promote COOperative research. On the other hand,
We should franklyrecogaize the difficulties involved. State lines
seem to be realities of a most stubborn-nature when it comes to getting
tOgether in affairs involving research and educational activities. One
man says: ' '

"The individuality of workers in institutions is too great. ‘
There is a frequently sub—conscious ”attitude of hostility in the, 1
way of. direction or coordination in a department."

This section of the handbook c-an‘consider no more important
subject than that of working out effective means of cooperating with .
one's neighbors in the deve10pment of long-time programs of research ~ '
which have to deal with problems common to several or many states in '
all of which commendable efforts are being made to study the economic
problems of the state's agriculture. The situation is not as hopeless
as might be inferred from the very limited success with '-it thus far ‘-
but it is nevertheless difficult. Cooperation is this as in all things
first has to be learned, and then it becames a habit. The conservatism

444

I

which seems to be manifested in many states is not entirely without
justification. Inevitably in a field so new as our own, many false
starts have been made. Ideas and policies in research have changed
rapidly. Too much of the research of the last ten years in this

the field.

One type of cooperative effort relates to regional surveys
preliminary to regional programs. The flepartment of Agriculture
has promoted or shared in nearly all of these. Some of them have
already yielded results in the way of publications. Others have not
yet reached that state. The correspondence contains a number of
references to'such work. rIhe following is quoted from a state leader
whose section is participating in the Northern Great Plains regional
study:

"The particular study which you refer to is the one
which has given me personally most Concern, namely, "A Study
of the Northern Great Plains Agriculture." I am. not at the
present time satisfied with the organization of this project
and I think that this is pretty much the judgnent of the
other states concerned. I have stated that it seems to me
this is partly a. project in studying how to organize and ad-
minister regional projects, and we have a number of details
to work out in that connection......... The difficulties,
however, are not from undue dictation by the Federal Depart-
ment—- if anything it might be considered a little bit the
other We . I think we might put our finger on the trouble
by saying that no one person is sufficiently reaponsible in
this project and the duties of the various individuals are
not sufficiently clearly defined. Also, some of the mech—
anical details of administration need still to be worked
out.

"This project was first taken up as a. reconnaissance
survey and in ,the conference following this survey the plans
for the long—time project were developed. We have no criti~
cism of the Opportunity given the various states toexpress
their views of the Various problems. It is just possible that
there are too many people concerned in this particular project
in that it involves both the farm management and the animal
husbandry sections of the six cooperating agencies, which means
.twelve people besides the station directors to be censidered."

This particular comment stresses the point already em-
phasized, namely, that in any such cooperative program the first
essential is able leadershipwithhsuch leadership well defined and
understood by all the parties. '

Another of the state leaders has this to say with reference
to the same projectzg’

445

"It is my opinion that the Bureau at Washington has
greatly improved its methods of procedure and its relation-
ships of'late; in fact the work as now conducted makes pos—
sible a greater volume of output than formerly and has the
adVantage of pooling the brains of the Department with those
of the States, which is desirable in my cpinion.§

Following is a report from another state on'a regional study
of cooperative elevators:

"This project has worked out, I believe, very wall; Ibch
state has attacked the general problem and taken up one or two
specific problems for intensive study :dthin its state. I be-
lieve there have been annual neetings of the research workers
on the project, and I know we feel that as far as our state is
concerned it has been very successful and the interstate and
department relationships have given us far more returns for our
research dollar than Would have been the case had we been going
it alone. It has been my observation that the principal in—
terest of some states in the work of the Bureau is to get money
out of them to supplement their own state funds, and that after
agreements are drawn up supplying this money, these states some-
what resent further supervisional activities on the par of the
Bureau. It is my feeling that both the work of the Bureau and
of the states has passed the promotion stage and that cooPerative
projects should be based primarily on a frank recognition of the
contribution which each party is in the best position to make."

Another state man discussing cooperative relations with the
federal Department of Agriculture says:

"In the formulation of projects there is free discussion
and c00peration between the leaders of the project representing
the various interests concerned. Frequently the man in charge
of the project from the United States Department office spends
two or three days or a week in the state in consultation with
the state project leader. Together they work out the details
of the projects and prepare the final project statement. In'
some instances the leader of the project in the state visits
the Washington'office for the purpose of formulating a pro-
ject. Occasionaly the project is formulated without a pee-
sonal conference. In such cases there is free correspondence,
criticism of project proposals, and reconstruction."

Also the following:

"we have had both cases, those of projects worked out
in conference with state workers and projects formulated by
the Department and presented as finished plans to the state.
I believe both of these plans have their place, though, of

446

.J.
course, in a majority of cases the plan of conference with
the state workers is to bepreferred, I think. The Depart-
ment can be of great assistance by serving as the means of
bringing together workers in a given region for the working
out of both projects. It is somewhat awkward and difficult
for one state to take the initiatite in this connection unless
it can do so through the Department.

"I would say that it would seem to me that the states
having in mind regional projects should communicate with the
Department and possibly at the same time with other proppec—
tive cooperating states. The Department might well then feel
out of the sentiment of the states which should logically be
concerned and, if this sentiment proved favorable, arrange for
a meeting to plan the regional project. The direct supervision
of the project should at that time be Specifically planed either
with the Department or with some one of the representatives of
the. cooperating states. "

An important phase of project formilation is the embodiment of
the essentials of the agreement in a formal project statement which
stands as a sort of contract between the cooperating parties. The
purpose of this statement is, of course, to serve as a memorandum of
general objectives of the project and of the procedure which is to be
followed in the carrying out of it. It has been the custom in the
past to pay a considerable amount of attention to the drafting of this
statement. Doubiless this is wise, inasmuch as it leads to more care-
ful consideration of the fundamental Considerations lying back of the
statement.

A very significant movement in the direction of cooperation
between state agricultural experiment stations and privage and semi-
private organizations has been developing in the last few years. The
more conspicuous work along this line has centered around the relations
of eatperiment stations to farmsrskorganigations such as the Farm
Bureau, the Grange, and the cooperatives or federations of such coopera-
tives. There has also been a limited amount of research work developed
between experiment stations and private comercial business. In many
cases these cooperative relations have been extremely helpful, but in
some other cases danger has appeared that these relations may lead to
embarrasment and to injury to the long—time interests of research. Ob-‘a'
viously very much of the success or failure of these relations depends
upon the cooperative procedure which is employed in working out the
joint program with such organizations. One or two basic considerations
should continually be kep in mind. The first of these is that none of
these organizations exists primarily for research. This is not their
Special field of endeavor. It is only in exceptional cases that; anyone
in such organizations has an adaquate point of view on research.‘ It is
therefore important that the leadership and guidance of such problems be

44’?

retained by the officials of the experiment station. The other
point is that the experiment statiOn must guard itself against any
charge of lending its effort to propaganda. for special interests.
That this is not easy is proved by recent experience with the power
interests of the country. '

Elie following comments from corr68pondents bear on the fore-
going observations:

"We have done a large amount of informal cooperation
here. The Farm Bureau Committee on Marketing goes eyer the
work of the dspar inert once a year all-:1 makes recozmendati ens
concerning the research work in the fie} d?!- in which they are
interested. 57.71311 recommendations are welcomes so heartily
that farmers and. formers’ organizations have no hesitancy in
consulting and recommending. Much of our research Work is
planned after such consultations.

Also the following from a leader in one of the northwestern

states: ‘

"We conducted two marketing surveys in this 'state on
which We relied on cooperation from the Commercial Clubs. We
asked thme to Contriouto b0 th clerical wcrl: and enough money
to show they were vitally interested. The chief asset in
this experience was that it gave us a such better working
basis with wholesalers and retailers than ifwe had tackled
the job alone. '79 had. the Connercial Club appoint a committee
which was made up. principally of broadmir.o‘..ed wholesalers and
retailers, who were .h:.t«'~;i".ligent enough to lmow what we were
after and that we were not spying and snooping around dark
corners with the idea of bringing out some radical revolution—
ary program. In each step in this procedure we held a con-
ference With the cormittee and we would wait; a week or ten
days before following up the line of work which was discussed
by this committee. By so doing, We had coqlete access to the
accounts and records of a number of firms that would never
have given us anything but a pleasant smile had it not been
for this action.". '

Another leader apparently has had a somewhat different
experience. He Writes as follows:

"It is not so satisfactory in my judgment to tie up
with commercial organizations or with state political depart-
ments.-~—--—There always is a danger that the leaden of the
commercial organization'will try to gain a reputation for
himself or that the state department will try to capitalize
the undertaking in a. publicity my long before the time for
definite conclusions has been reached."

448

Those who answered inquiries on this'subject had some rather
definite suggestions as to the procedure in develOping research methods
on a single project or in a pregram. Dr. Warren took the individual
rather than the c00perative side of the question:

"Most good research work is personal rather than committee
work. Occasionally two men naturally work. together and ac—
complish more than either one could alone, but ordinarily the
best way to get research work done is to chooses. good man let
him alone, giving him only such advice as he asks for. I do
not favor trying to have work in different areas too much alike.....
I "think each man should be encouraged to use his own ingenuity
and carry on as complete and extensive a study as he is capable
of doing rather than take any stereotype fonn and try to use it
everywhere. I cannot Conceive of a good research man being
willing to allow anyone else to completely dominate his work,
altho any good researdh man nearly always welcomes suggestions
from any one."

Here again the-necessarily close relation between the federal
Bureau of Agricultural Economics and the various states is stressed by
a number of men. Professor Grimes of Kansas says:

The federal Bureau (can perform real service in helping
to develOp new and better methods of research. This is par—
ticularly true in the farm management field Ifnere we are very
much in need of a less expensive method to replace our statis-.
tical routes. When a problem of developing method is inclosed,
the federal Bureau is warranted in cooperating on problems that
are distinctly local in charanrter and pertain to regions entirely
within the state.“

The secretary of one of the research'councils says:

"Methods of research are discussed with the secretary
of the council from time to time and are also presented by
Specialists of the Bureau at annual meetings."

Another federal Worker states as one of the purposes of
the research council movement:

"To. insure desired comparability in the collection and
subsequent interpretation of research data, the Bureau is by
nature of its national aspect“ in position to aid in bringing
about the desirable end."

Most research workers would agree with Dr. Warren that there is grave

danger to the progress of research in too much comparibility along
these lines.

449

Further comment on the function of the Bureau in this con—
nection by a state man is as follows:

" In developing methods of research there have been
joint agreement on most lines. On turo or three occasions
the men in‘ charge cf the research work in the state have been
called to Washington at the expense of the Bureau to sit in a
tab or three day conference on methods of securing data, in
the interpretation of data, and in the developing plans cover-
ing methods of field work and of handling data. "

‘ Another state manjsays:

"The work which is being done in deve10ping methods of
research will, Ilthink', be the ”most helpful, at least to us
here. l have a feeling that in an experiment stations there
is considerable wastage’througi use of'inade'quate methods.

. To sum up on the matter of cooperative procedure in develop—
ing research methods, it would seem that the first and most important
consideration here, as in other phases of our general problem, is that
of bringing into effective functioning the best leadership that is '
available in the field. Where is every reason for keeping all workers
fully informed as to the methods that have proved most effective in the
hands of our best thinkers; but there is, grave danger that ill—Considered
imitation may appear in the form of applying new and promising methods
by inexperienced men to problems for which such methods are not adapted.
COOperative effort in the form of frequent. conferences is no doubt to_
be encouraged; but the task of developing‘method is more largely an in-
dividual. rather than. a cooperative one. Method results very largely
fromexperience, and-experience develops primarily in the form of in—
dividual effort. It is probably true that practicallylall of the im-
portant advances in research technique in our field during the last
ten years have come as the result of a long period of eXperimentation
and study on the part of a few individual workers who perhaps from
time to time have compared notes formally or informally. Impemw
the development of a” given technique, such as the application of cor-
relation methods to economic"pro'_blems, has been'givan through the
publication from time to_ timef‘of' pioneer pieces of work.

'One specific proposal along this line is for the exchange
of results among research workers to contribution. This proposal
has some ticklish aspects, inasmuch as research workers are sometimes
overs-sentive and over—anxious in the matter of securing ade'quate'credit
for new. ideas. HoweVer, this should not stand in the way of the more
rapid progress of research which should come from a free exchange of .
,ideas in the 'Various working stages. of the Carrying out of a. project.
This is pointed out admirably in the reply to one of the writer’s
letter of inquiry, "I believe that there" is great need for making .
available to research workers unpublished manuscripts which may re— «L95

450

port careful work and certain unpublished materials which might be
of interest to other workers."

3. In carrying-'J'I‘oit ' As a usual thing the cooperative pro-'

joint 93 parallel cedure in carrying out projects is Forked
projects. . . . ,out at least in its general‘outlines and

specifically noted in the project‘state—

. _ ‘ ment. ‘At best, however, these statements
are somewhat general in nature and there is always the further prob-
lem of the detailed workingout of cooperative projects in terms of
just how much time this or that worker. shall devote to it and who
shall havethe‘ work (of analysis and. of composition looking toward pub-
lication of results. On cooperative projects with the Bureau of Agri-
cultural Economics where field work is involved.‘ These arrangements
have been well worked out and need little disaussion. Where a number
of field men are employed, some sort of supervision is needed. ’ This
is sometimes supplied by the Federal Bureau and sometimes by the states,
depending very largely upon the extent to which the-state has already
developed research work and has adequate leadership. - Cooperative pro-
cedure in the analysis of data and its preparation for publishing is
not, however, so satisfactorily developed, as evidented by-the follow-
ing from a letter“ from a state leader: > ' " ' '

"Where projects are conducted couperativ'ely by the
federal Bureau and‘ezperiment stations, much of the data
has been analyzed in Hashingtom. . As a consequence the
state workers have lost all contact with the analysis of '-
the data, ' In my opinion this is wrong, as the state
workers lose contact with the project and do not get the ,
benefits accruing through having'to'fatte the Various prob-
lems that are bound to arise in the analysis of the data.
Some arrangement should be perfected thereby representa-
tives of the states and of the federal Bureau could work
on the analysis. It muld-‘be equally as undesirable to
have the states do all the analyzing to the exclusion of
the workers from the federal 'Bureau. 'fliis point presents
problems that. are not easy of, solutiOnu,‘ particularly where
a state has limited personnel and. theithime- is divided
between teaching, extension, and research." However, it
is believed that a. much more satisfactory arrangement could
be arrived at than has been followed in the past where the
data for many projects have been collected Jointly and then
analyzed in Washington to the exclusion of the state workers."

No doubt the cooperativeoarrangement worked out for the analysis
of the data has been responsible for unfortunate delays-in publication.
Both the federal Department and Various of . the states have doubtless been
at fault in this conditiOn. Many of the states have inadequate pro-'- ~
vision for clerical help, and their professional staff is held for too
heavy a. load. The writer has in mind a project carried on cooperatively-e
by a state and two Bureaus of the federal Department, the suggestionfor

the cooperative work comingf'fron the Department a": lgrictu ture, The

451

field wail: imlved 3000 farmers which meant the work of a crew of six
men through aperiod of about eight weeks. The field schedules were
taken to the repartment of Agriculture for smalysis. ‘There developed
some difference of opinion between the tWo Bureausof the Department

as to the feasibility of publication. The result is that this rather
heavy investment which was made in 1923 has thus far yielded no tangible
results either for the state or for the Department. Undoubtedly con-
dit ions will arise under the best of cooperative organization which
will eeus e occasional miscarriages of this sort. Coo: erative procedure
in the future, however, should look toward the elimination of them as
far as possible.

Perhapsthe' best way that has yet been hit upon for'the an—
alysis of material obtained from cooperative research projects is that
which has been developed in recent years by the federal Bureau. It is
that of sending into the state, upon agreement with the state author-
ities, a. federal worker who will stay in residence for some weeks or
months cooperating closely with the state men in the analysis and pre-
paration for publication of the results of the joint project, There
“is little doubt but that the resulting publiCation has been- stronger
' and that the beneiits accruing to both of the coo )ere ting parties are
vastly greater than if one or the other had undertaken and carried
through the preparation alone. The alternative arrangement, however,
of having the state leader work on the project for a period in Wash—
ington, has as much in its favor.

4. The UtiliZation The general aspects of the problem
of the Been—its of utilization of the results of research
flim- will be handled in a later section. Let

suffice here a few remarks on the cooperar

tive aspects of it. The most significant
example in this sphere of activity in late years has been the joint of;
‘ fort in promulgating the federal outlook report. The details of this
are too familiar to need discussion. Each ye r the states are taking
hold of this extension project in more earnest. The federal Bureau
needs more men when it can send out to participate in the state and
regional conferences_in which the general program is adapted to local
conditions and expanded into more detail as is possible for a smaller
area. The regional extension programs that have been worked out at
conferences between state and federal extension workers with research
and subject—matter men aesisting in council, represent the cooperative
mode of attack on the more long-time aspects of the problem. The
federal Bureau's sending out representatives to sit with those of
local departments of economics in important local conferences, is
another type of cooperation in this field.

452

Dr. E. W. Allen, Chief of the Office of Experiment Stations,
was asked to review this discussion of cooperative procedure. .Most of
his comments have been incorporated in the text; but the following need
quoting exactly: "Cooperation is only promising and profitable among
those who believe in it and are willing to join hands in a cooperative
spirit. A cooperative enterprise seeks to bring these together; some
will never conform, either in letter or spirit. They are better off
outside."—--—~ "Thus far our cooperation is not very effective. Uh-
leSs there is a plan with definite purposes to be served by a confer—
ence, it will become mainly a talk-fest. Talking things over will
not suffice. Something should be crystalized out of a conference which
will make the cooperation more effective."—--~- "Cooperation in re;
search is not designed to stifle individual initiative or subordinate
workers to mere routine. Participants oughTto have an individual part
in the research - not to be working for some one nho is to do the re-
search and interpretation."

 

 

Overhead organization* The subject of econondc organ-
gg affecting research ization of the land — grant institup
in agricultural economics: tions in the broad sense of departs
1. Scheme of organization. mentalization and lines of authority

and responsibility, and interrelap
tionships between_divisions, is one meriting careful-analysis. But it
is out-ofhplace here except as it relates to the research in agricultural
economics. All these institutions of course have three lines of work -
teaching,~research and extension; and it is proper to Say that these are
the three main functional departments in each institution. If for no
other reason, the sources of funds are sufficiently distinct so that
separation of the work along these lines becomes necessary; But here
the uniformity stops. Table I shows 5 different schemes of overhead
direction of these three fields of work. First are 8 states in which
one person is head-of all three lines of work. This really means that
the head officer of the institution, usually called "dean and director"
in such cases handles the major administrative details of cash of these
lines of work. He probably has an assistant dean of the college, and
an assistant or vice director of extension and of research to look after
smaller adminiStrative details. There are twenty in which a separate
person is called "director of extension" and handles all the details
of administration of the institution, subject, of course, to general.
supervision of the superior officer of the institution who.is usually
either a president or the dean and director of the other two fields.
In 3 institutions it is the dean of the college who is given this
more distinct status; in 3, the director of the station. Finally, in
L4 institutions, all three divisions have distinct heads, who of course
are subject to the superior officer of the institution, a president or

 

* By G. ?. Forster, North Carolina Agricultural College.

453

.Ta'ble ,1. - Relation of Overhead Organization to combination of

 

 

 

 

 

 

 

 

 

Functions in Economics Departments.
: :Not
Overhead Organization :States in Ylhich Economics Depart- :Classed

:’ ' :ments Handle Different Combinations:
Nature of iName of State :of Functions :
Organization z : : :' : :
: =Res.Teaching :Res.&' :Res.& :Res.:

1. _ : = Extension :Teach. :Egn‘fi. : : _

ne Office}; : Alk.,Del.,F1a., ‘_ :Fla.*111., Ky." IArkJ.‘ lFla.’ my: Ida.

Deer: of Coll. : Ida. ,Ill. ,Ky. , mils. mel. : : :
Dingpf Sta., : Penna” Wis. : :Fla. : :
end‘Lir. of Ext.: 3 1190111151.: : :

Tm Office_rs_ :‘Ala. ,Ariz. ,Calif. , :Ala. ,Iowa, : Calif:Vt. : :kia.“
(l)Dea.n of Coll. :Iova,l§ans. ,I.a. , :l-Zd.21-£ass.*tio. , :Minnu": : : . H
and Dir.of Sta. :Md. ,Mass. ,2.[5.cn. ,liinn Mont. ,Nebr. ,N.J. :Okla. : :i‘uiich.

:Mo. ,Liontaflebr. ,N.J. :S.C. , Wash. :17. Va. : : :
(a)Dir. of mt. :Ol:la.,S.G. ,Vt. [El-3.811.: :Tiyo. : : :
:I’Jfla. £330, : :Kz-n. : : :
Two Officers : ‘: ‘ : : : :
(l)Dir.of St. :Ind., N. H. , Utah :Ind., N. H. :Utah : : :
end Dinof Ext. : : : : : :
(2 )Dean of Coll. : . z : :
Two Officers : : : : :
(l)Dean of Coll. : Me., N.C., R.I. : :Me. = : : R.I.***
and Dir. of Ext.: : :N.C. : :
(2)Dir. of Sta. : : :4 : :
Ehree foi.9_e_r_§ :Colo. ,Conn.Ga. ,Miss..: : : : :
(1) Dean of Coll. :Nev. ,N.Liex. ,N.Y. :Colo. ,Conn. , :N.Dak“ :N.DakflMisszGa.***
(3) Dinof Ext. :N.Dal{. ,Ohi090re. , :N.Y. ,Ohio, :S.Dak.: : :Nev.
(3) Dinof Research:S.Dak., Tenn.Texa8 :Va. : z : :N.Mexicc
fife. : - : V : : '3Texas*
*Separate departments of Agricultural "No Research "“No Department

Economics and Form magem‘ent

454

dean, who is consulted before any major decision is made. Even this
classifiCation'does 'not'pro'vide for all the differences in detail.
Minnesota, for czample, has a dean of the college and a director of

short courses, - Many of the above differences are also merely paper

mleifferences. A person called director or dean in one institution may
have no more responsible position than an assistant director or dean
in another.

. . The second part of Table I attempts to show the functions
,. assigned to the economics departments and whether there is any connec-
tion between this and the overhead set-up. It appears that there are
2;. economics departments which handle the work in the three lines,
teaching, research and extension, subject to the general direction of
the deans and directors. This direction is necessarily much more
detailed in the case of the extension work, in many cases covering
nearly-everything exeept the supplying of the subject-matter to be
extended. It Will be noted that there are only two other states in
' which economics departments handle extension work, Vermont and North
Dakota. This means that in the remaining 16 states in the second
column the extension division has its own staff for the economics
work. There are 16 other departments which have research and teach-
ing combined. All these numbers count farm management, marketing and
the like as separate departments where they are so organized. In 36
states the economic research is all in one department. Only three
stations have economic research in a separate department from the
teaching. From the foregoing, it would appear that the peculiar
organization of institutions has had little or no effect on the way
in which the department of agricultural economics is organized. In
those states which have but one executive officer in charge of all
lines of activities, the agricultural economics department in four is
administering teaching, research and extension. In those institutions
which have two officers, there were 12 in which the department of agri-
cultural economics administered research, teaching and extension.

In institutions in which one dean or director is directing
all three lines of work, and the economics department handles all
three functions, the problem of lines of authority and responsibility
is very simple. The only question is when to call upon the assistant
dean or director in each field and when to call upon the chief officer.
Probably most matters g) first to the assistants. In four cases,
AranBas, Delaware, one of the economics departments in Florida, and
Pennyslxrania, the extension work is outside of the economics department.
Then the anninistrative problem is to keep the economics extension in
line with the teaching and research; but since the chief officer has
all‘three under his direction, this can no doubt be accomplished, al—
though it may prove difficult.

In the 14 institutions where there are three separate func-
tional heads, the 5 departments of economics having all three lines’of

~

455

work have to keep clearly in mind the difference, between the three 'fimc-—
tions and the relations between them mdrknow when to call upon each
head. lhe analysis vfnich this entails is extremely valuable from an
organization standpoint. The three coordinate heads probably watch
closely how well their functions are being performed within each depart-
ment. In 4 of these 14 institutions, the extension department has its
own staff in economics, and in another case extension is combined with
research. With a separate extension directors-ith an extension staff
in economics of his own, the difficulty of keeping the two in line and
the extension work on a high plane so f or as subject—matter is concerned,
hasproved will weigh msurmuntable in the usual practice.

Any kind of a two-officer arrangement is some kind ‘of a combine;-
tixgn of the two foregoing. In only 9 of the 26 states with such separate
economics staff, and in 3 of' these states, the direction of extension is
also director of research or, dean of the college.

In the few cases in which research and extension in economics
are in separate departments, the ranking member of the staff is likely
to be only a sort of nominal head, most details being handled directly
between the staff members and the director. A condition resembling
this often maintains even when research and extension are in the same
department with teaching. The military type of organization in which
the staff members deal only with the head of the department, and the
department head only with one superior officer, or one for each line-
of work, is not carried out rigorously in most institutions.

The various members of the staff may in fact not negotiate
many sorts of details with the head of the department, but with one
or more ofthe three functional executives, depending upon the nature
of the work. That is to say, in case a staff member is conducting
research he may go directly to the director of the eXperiment station
for consultation; or on the other hand, if he is engaged in teaching
he may go directly to the dean of‘instruction, and if engaged in ex-
tension work he may be responsible to the director of extension for
many sepectsof his work. It is always assumed that the department
head is informed as to what is going on.

No doubt the type of organiZation affects the research work
in mny ways which cannot very well be reduced to chart or printed
form. . There are certain evident advantages of an organization which
has one all executive officer who has achieved a standing in the educa-
tional world. Such an officer can exert a greater influence on the
nature of research and also coordinate it to- a better advantage than
two or more executives all interested in promoting their own work
perhaps without due regard to other but related types of work.

As to the matter of functional- relationships and departmental
organization, one of the deans who has had wide experience has this
to say:

456

"I am strongly of the Opinion that the closely knit
‘type of organization is the most desirable. I have worked
under both forms in different institutions and certainly feel
' that no subject—matter department can be a. functioning depart--
ment unless it has some stronger tie knitting it together than
the tie of subject-matter. It is my considered judgment that
most subject-matter departments headed by a. chairman who gets
all of his salary from one subdivision simply cannot function
in the best possible way, for where the treasure is, there is
the heart also." '

The type of organisation of the research agency probably af-
facts also the departmental program of research. The recent survey of
the economic research in agriculture conducted by the Social Science
Research Council showed that it was uncommon to fund a unified program
of research covering the entire field of agricultural economics and in-
floating the relationship of the various projects and functions to each
other. While this might not be traced entirely to the type of organ-
ization, yet. there is reason to believe that a closely knit organization
would realize the need for a comprehensive program. The type of organ-
ization no doubt affects the relationship between research in economics
and other fields.

The foregoing discussion has been entirely in terms of land-
grant institutions. The same principles will apply to any other research
organization that is departmentalized.

2. Allocation 91‘, funds It has been recognized for some
1:2 economics. resegch time that agricultural economics research

has not beenedequately financed. Unfor-

tmately the land-grant institutions while

in many instances realizing the importance
of this type of work, were not in a. position to emport it financially.
One difficulty arose with the limitation placed by the federal administra-
tion upon the federal funds made available by the original acts establish—
ing the land-grant colleges. The Hatch and Adams Acts were for the most
part interpreted as referring to natural science research. The stations
have therefore had to depend upon state funds or other funds for the pro-
motion and development of agricultural economics research. And. as these
funds have been rather meagre, the growth ,of the work in this field has
been seriously retarded. Lack ‘of properly trained personnel has of course
been another limiting factor; but personnel tends to develop as funds be-
come available. Fortunately, however, the Purnell Act passed by Congress
in 1925 has made available funds which can be used wholly for economics
and sociologfitfit'g‘tes so desire. One barrier, then, which has so long
interfered wit the development of this work has been removed.

457

As the Purnell fund has been available for the past two years,
it is interesting to note its effect upon the amount of funds being de-
voted to agricultural economics research. In 1924—5, the year prior
to the pamage of this Act, the state institutions Spent $355,859 in this
field. Most of this caste at of state budgets or out of the budget of
the Bureau of'Ag-Dricultural Economics. In 1925—63 the states spent $608,000
for agricultural eCOnomics research, an increase of $352,161 or of 70.8
percent. rI'he total increase in federal funds received Was $906,000. Of
this, $398,838, or 74: percent, except the small amount that went to home
economics and. rural sociology, went to old—line research or Was otherwise
absorbed in institution budgets. In the year following, the funds for
this field were further augmented, the total being; $758,781, or an increase
of 116.0 percent over 1921-5, and 26.4 percent over 1935—6. fliedncrease- ’
infiederel funds to the 50 institutions Was $480,000,0f which $311,219, or
65 percent went elsevmere than to agricultural economics research.

One of the ways that the increase in federal funds is absorbed
appears in Table II, which shows that during the period from 1924-5 to
1926-7 there Was a diseppeamence of state or other federal and miscellan-
eous funds of $541,745; that is, when the Rimell funds actually spent in
1926—7 are added to the amount Spent in 1924—5, the total is $833,527 as
against the actual outlay of $768,732. From this it mind appear that
state support to agricultural ecorcmics research has been withdrawn as
the Pumell funds have been increased. The actual displacement has been
much more than these figures indiCate. In Ohio, for example, a Special
state appropriation for research in agricultural economics swelled the
total of state support. In institutions where teaching and research
are done in the Same department, the staff has had more time for re-
search with the reduced undergraduate enrollments since 1925; and this
has resulted in an apparent increased allocation of elependiture of state
funds to research, although state support has been actually decreased.
One of the most obvious ways in which state support is replaced by
Purnell support is to pay the salary increases of those doing both
research and teaching entirely out of Purnell funds. But state sup-
port of agricultural economics budgets has been actually reduced absol—

. utely in many states. Up to 1328, nearly two thousand dollars of state .
support had been taken away at Minnesota. Other states can show a. worse
record.

rfable III classifies the states according to the amount of funds

spent on agricultural economics research in 1925-7. The picture is some-
what confused by the fact that there are two departments of agricultural
I economics in some states. In Spite of the passage of the Purnell Act
' ' giving federal funds and to all state regardless of size, there is still

‘ a decidedly unequal distribution of financial support for agricultural
economics research. Comparison of the last two columns in Tables xv and
v shows that change have occurred in this respect. First of all, there
was an increase from 35 to 46 states doing research in this field. In
1924—5, 3&3 percent of the states spent 8.0 percent of all the fluids
spent in the 48 institutions on agricultural economics research; in

458

Table 2.- Displacement of State Support by Purnell Funds in
Agricultural Economics Research from 1924-5 to 1925-6

 

 

.1. .2. .4. -4- .5. .6. .7.
. Funds in P11111011 Total Funds Decrease or Increase
No. State 1924-5 1925-«7, 3 and 4 1925!? In 111p£airom 517.5112.
1 Ala. - 7,400 7,400 7,500 {r 100
2. Ariz .
3 Ar; 1,100 ' 18,450 19,550 19,550 -
,.4 _ Calif. 20,000 15,892 35,892 27,000 —8,892
5 ' 0010. 10,000 12,200 22,200 13,700 -8,500
6 ' Conn. 6,500 16,000 2.2, 500 ‘ 16,000 45,500
7 Del. — 5,800 5,800 5,800 -
8, ’ Fla. - 16,050 16,050 16,050 — ,
9 Ga. 4,000 —- 4,000 4,200 «5 200
10 .Ida. - 5,594 5,594 5,807 +. 215
11... 111. 25,855 - 25,955 47,818 55,985 +-5,157
12 z;d. 9,887 5,800 15,587 15,500 +-1,915
15 ’IQWa 52,800 5,590 58,590 40,240 +-1,85O
14 Kane. 1 6,500 12 , 891 29 ,391 16, 450 ~12, 941
15 Ky. 10,190 17,740 27,950 18,555 - 9,555
16... La. .. - .. .. -
17 Me. . ~ 11,500 11,500 11,500 -
18 M1. 1,200 5,000 7,200 7,800 1. 600
19 Mass. 3,320 12,840 16,160 12,990 - 3,170
.20 Mich. 5,500 15,000 20,500 20,530 - 400
21 Minn 24,189 12, 5,213 35,712 40‘ , 859 + 4, 157
22 Miss 5,000 6,25 11,250 6,250 - 5,000
25 M0. 9,100 4,480 15,550 12,880 — 700
24 Mont. 7,480 8,100 15,580 19,000 .4 3,420
25 nebr. 7,700 8,555 15,255 14,550 — 1,925
26 Nev. — 11,200 11,200 11,200 -
27 . N. H. 9,000 8,761 9,661 8,761 - 900
28 ' N. J. 5,500 8,040 15,540 14,020 15 480
29 N. Mex. - 8,700 8,700 8,700 -
30 N. Y. 47,000 13,400 60,400 60,400 ..
31 'N. 0. 4,000 5,383 9,383 7,183 .- 2,200
32 N. :0. 19,100 8,850 27,950 22,775 .- 5,175
55 Ohio 2,500 ' 11,550 15,850 .27,400 —+15,550
34 Okla. 4,100 12,400 16,500 13,300 - 3,200
55 “ Oreg. 2,000 11,500 15,500 14,125 1. 825
36".: Penna - 10,000 10,080 10,437 + 437
37 R. I. - 6,130 6,100 6,100 -
38 S. 0. 4,500 13,000 12500 15,300 — 2,000
39 S. Dak. 11,000 15,630 26,630 15,630 ~11,000
40 Tenn. — 2,000 2,000 2,450 - + 4:50
41 Texas 8 , 000 10,900 18 , 900 2'1, 950 ' + 3 , 050
42, uwm 4AM) zmm’ ugwo 118m +12%
43 Vt. 1,150 9,000 10,500 ‘ 9,000 — 1,520
44 Va. 8,000 15,000 23,000 15,000 — 8,000
45 wash. 2,555 9,214 11,57 9,814 _ 1,755
46 W. Va. , 4,345 6,500 10,845 7,250 , 45,595
47 Wise. “27,000 5 ,55 32, 550 36,720 + 4, 170
48 Wye. $500 11,200 , 12,700 . 11,202 - 1,500

 

355,839 ‘457,588' 825,527 768,782 ' »-54,745

.1

Table III. -

459

Classification of states according to the

amomt of funds used in agricultural

economics research in 1926—7.

 

Amount

State

0
t

No. of States

 

a .
Arizona, gap, Lg” Term.

 

 

Q. g. u. on o. .-

 

 

 

 

 

 

 

 

 

 

 

 

_L_e_s_§ then $ 5000 . 4
; Alan Del._, 1:13., Md., 153.55., N. H.
550000—9399 : N.‘Me::icoLN. 0., R. I., W. : 12
: Colo. , 1410., 11:335., 110., Nebr., Nev. :
510,000-15999 : N.J., 01:13., Ore., Pa” Utah, Wye. : 12
: Ark” Conn., Fla., Ind” Kane. Ky. :
@1000‘19,999 : Lionto 1 5.0.4 5.91:8» 1 V8.- : 10
304000441999 : Mich. , ILDalc. , Taxi; : 3
$25,000-29,999 : Calif” Ohio : 2
£30 1000—84 , 999 ; : O
$35,000—39J999 : Wis, : 1
$10.0CO~44,999 : Iova,ginn. : 2
$454000-49.999 ‘ : ': 0
3&0000-544999 Y ; Ill. ; 1
3554000491999 : F. o
$60,000-641999 : New York : 1
* No funds.

460

 

 

 

Table IV. The Distribution of Support for Agricultural
Economics Research, 1924-25. (35 states)
State : Funds : State
: : Cumulative
: 1924—5 Cumulative : Percent
: Percent :
No He 3 $ 900 5 025 : 2.78
AFR. : 1100 : .56 : 5.56
M3. : 1200 : .90 : 8.33
Vt. : 1500 : ~1.82 : 11.11
Wyn. : 1500 : 1.74 : 13.89
Ore. : 2000 : 2.30 : 16.67
Wash. : 2365 : 2.97 : 19.44
Ohio : 2500 : 3.57 : 22.32
Mass. : 3320 : 4.60 : 24.50
N. C. : 4000 : 5.73 : 27.78
Ga. : 4000 : 6.85 : 30.55
Okla. : 4100 : 8.00 : 33.33
Utah : 4100 : 9.16 : 36.11
W. Va. : 4345 : 10.38 : 38.89
S. C. : 4500 : 11.54 : 41.66
M158. : 5000 : 13.05 : 44.43
N. J. : 5500 : 14.59 : 47.82
Mich. : 5600 : 16.17 : 50.00
Conn. : 6500 : 17.99 : 53.78
Mont. : 7480 : 20.09 : 55.55
Nebr. : 7700 : 22.26 : 58.33
Texas : 8000 : 24.51 : 61.10
Va. : 8000 : 6.75 : 63.88
Me. : 9100 : 29.31 : 66.66
Ind. : 9887 : 33.99 : 69.44
0010. : 1000 : 34.90 : 72.22
. Ky. : 10190 : 37.76 : 75.00
S. Ink; : 11000 : 40.85 : 77.78
Kane. : 16500 : 45.49 : 80.55
N. de. : 19100 : 50.85 : 83.33
Calif. : 20000 : 56.48 : 86.10
Ill. : 23863 : 63.18 : 83.89
Minn. : 21189 : 59.97 : 91.66
Wis. : ZTCOO : 77.57 : 94.41
Iowa : 32800 : 86.78 : 97.23
N. Y. ' 47000 : 100.00 : 100.00

Table V.- The Distribtuion of quport for kgricultural
Economics Research, 1926—7.(46 states)

461

 

 

 

40869

: Eunds : States-
State : 1926-7 : Cumulative : Cumulative
: : Percent : Percent
Tenn. : $ 2450 : .32 : 2.17
Ga. : 4200 : .86 : 4.35
Del. : 5800 : 1.61 : 6.52
If§ho : ‘5807 : 2.3 : 8.69
R. I. : 6100 : 3.17 : 10.86
Miss. : 6250 : 3.92 : 13.04
N. C. : 7183 : 4.91 : 15.21
W. Va. : 7250 : 5.85 : 17.38
Ala. : 7500 : 6.83 : 19.56
Md. : 7600 : 7.84 : 21.73
N. MeX. .3 8700 : 6.90 : 23.90
N. H. : 8761 : 10.14 : 26.07
Vt. : 9000 : 11.28 : 28.25
Wash. : 9814 : 12.55 : 30.42
Pa. : 10437 : 13.92 : 32.59
Nev. : 11200 : 15.37 : 34.77
Wyo. : 11200 : 16.83 : 36.94
Maine : 11500 : 18.33 : 39.11
Mo. : M 12880 : 19.99 : 41.29
Masa. : 13990 : 21.69 : 43.46
Okla. : 13300 : 23.42 : 45.63
Utah : 13600 : 25.18 : 47.80
0010.. : 13700 ; 25.95 ; 49.98
N.J. : 14020 g 28.79 : 52.15
Ore. : 14125 : 30.62 : 54.32
Nebr. : 14330 : 32.49 : 56.50
Va. : 15000 : 34.44 : 58.67
S: G. : 15500 : 36.45 3 60.84
Indy : 15600 : 38.48 : 63.01
S. de. : 15630 : 40.51 : 65.19
Conn. : 16000 : 42.59 : 67.36
Fla. : 16050 : 44.68 : 69.54
Kane. : 16450 : 46.83 : 71.07
Ky. : 18565 : 49.23 : 73.88
'Mbnt. : 19000 : 51.70 : 76.05
Ark. : 19550 : 54.24 : 78.22
Mich. : 20200 : 56.87 : 80.40
TeXas : 21950 : 59.72 : 82.57
N.-luk; : 22775 : 52.68 : 84.74
Calif. : 27000 : 66.19 : 86.92
Ohio : 27400 : 69.75 : 89.09
Wise. : 36720 : 74.53 : 91.26
Iowa : 40240 : 79.75 : 93.44
Minn. : : :' 95.61

462

1926-7, this same percentage of the states spent about 14 percent of the
total. For half of the states, the comparison at the tile periods Would
be 16 and 27 percent; for two—thirds of the states, 29 and 41 percent;

" for 80 percent of the states, 45 and 57 percent. There is still a
pronounced concentration in a few states; and a pronounced opposite of
this in many others.

Tables VI and VII introduce some other aspects of the problem.
First, in 1926-7, in thirty states the economics research Was receiving
'70 percent or over of support out of the Purnell funds. Second, in
this year the ecjonomics research in' thirty-one states Was receiving 40
percent or less 'of the total Purnell funds available. If further ex— ‘
pension is to be‘made, it 'must apparently be out of the Purnell funds.
Exit further expansion from the Pumell fund- is becoming difficult because
a large part of the funds have already gone into other lines of work.
There is still time, however, to save some of the remaining increases;

Administration-,.* One difficulty. in the selection of-the
l. glection of research personnel arises, out of the fact-
reseagch staff. “that the persons finally responsible for

selecting it are not acquainted with the

field of'agricultdral economics. With new
departments and new lines of work being established, the dean or director
becomes largely responsible for the selection of the personnel. Unless
he has had spec‘ial training in economics, or has broadened his outlcSOk
by special study, he is not always in‘a position to pass judgment on
the fitness of a candidate for the head of a department or as a member'
of the department's staff. .In the past, most of the workers chosen
were men who had been trained primarily for some other field than that
of economics. And even now, many men are chosen for the research staff
who.are not, accurately speaking, economists.

Aside from this difficulty of obtaining qualified persons to
fill available positions, the selection of the research Workers is af—
fected by the research program. Without a definite program the selec-
tion of workers is likely to be haphazard. This indiscriminate adding
of workers will discredit, and no doubt has discredited the work in this
field. A well developed program of research would indicate which prob--
lems were more important in the larger run and hence indicate the type
of man that should be added to the staff.

. In some of our land-grant colleges, there is a decided tendency
to select the research personnel from graduates of the same department.
In some institutions, it has been the practice for the department of agri-
_ cultural economics to look over the records of Imdergraduates and offer
_ part-time work to several of the best of these each year. This gives the

 

*By G. W. Forster.

463

 

 

 

 

 

 

 

 

 

 

 

 

 

Table VI.— Classification of states according to the
percent the Purnell support is of the
total support provided for agricul-
tural economics research, 1926-7.
Percent : ,
Purnell is : : No. of
of Total : States -
Funds ; : States
1926-7. : :
Less than 10 : Arizona,“ G5,." La.‘ 1 3
10—19.9 : . Iowa, ‘Wi sconsin : 2
ISO-29.9 : Ind.., N. Y. : 2
30.39.29 : Minuet” N. oak. : 3
40-493 : 111;, Mont., Ohio, Texas : 4
50—59.9 : Calif., Nebr., N. J., Utah : 4
60,69.9 : : ‘0
70,79.9 : Kans., McL, Mich” N. C. 3 4
80—89.9 : 0019., Ore., S. C., Tenn., W. Va. : 5
: Ala., Ark” Conn.,Del., Fla., Ida., 'Ky. =
90-100 S: Mm, Mass., Miss” Nev., N.H., N. Mex., : 21
: Okla.., Pa.., R.I., S.Dak., Vt.,Va., Wash. :
: Wyo. :
Total 48

*No funds

464

Table VIL- Classification of the states according to the
percentage of the Purnell fund available
for 1926-7, spent on agricultural
economics research.(1)

 

 

 

 

 

 

 

 

 

 

 

 

 

Percent : : No. of
of Total : State :
Parnell : : States
Available : :
Less than 10 : 'Ariz.*. Ga.Lfi;e., Tenn. : 4
10-1939 : D61.) Idaol Ind. ’ 10%) MO. , N00. .1718. : 7
3 Ala... Ill. , hide, ILI‘IiSS. , Lion-tn , NQHI , .NOJQ , :
20-29. 9 : N‘Dak. I R. I. ) Utall , 1"].Va. : ll
: Me., Nebr., Nev., N.Mex., Ohio, 0re., Pa., :
30.303 : Texas” Vt. , Wash. . Wyo. : 11
: :Colo., Kane., Maes., Minn., N.Y., Okla., :
40—49. 9 : -S,C, : 7
50—59. 9 : Calif. , 00an g Fla. 3 KY. 1 MiChq g $190k. Va. : 7
60—69. 9 = Ark. : 1
70—79, 9 : z 0
80—89.9 : : 0
90.100 : ' : 0
Total 48

(1) Calculated on the basis of $30,000 per state
available for each state

* No funie at all

465

head of the department an opportunity to see how these men work at def.
inite task and also gives the student an opportunity to learn more of
the work of the department. After the student graduates from the
university, he may be offered an assistantship or instructorship. Then
a position of assistant professor is to be filled, it may be filled by
further selection annng these assistants or instructors. This policy
has certain obvious advantages, but it should not be pushed too far.
The dangers from it are relieved somefihat in a case of large institu-
tions where a great many of the graduate students are drawn from differ-
ent parts of the country and from other institutions. But even here,
it produces narrowness and lack of vision if carried too far. In the
case of smaller institutions, where the individual does not get an op—
portunity to work under a number of able men, the practice is at its
worst.

2. Erovision for A survey of the states reveals that
training graduate seventeen out of thirty—eight states re~
students. porting do not provide any financial

assistance Whatsoever. Fellowships and

scholarships are scarce; only eleven states
report scholarships or fellowships for graduate students. Assistantships
were available in ten of the institutions reporting. Many of these as-
sistantships were for part—time graduate students. There are very few
stipends which permit the individuals to devote all of their time to
graduate work. In a large number of the institutions, there is little
opportunity for graduate study; because the institutions are small, the
departments of economics and agricultural economics are relatively weak
and the number of graduates is so few that it is not worth While to de-
velop anything in the nature of a graduate course. Several of the states
reported such things as weekly meetings to discuss problems in the field;
others stated that they urged their assiStants and research Workers to
take graduate work at other institutions, usually without pay or any kind
of compensation. Others urged more study, etc., While the last thing
that should be urged is that each department of economics should attempt
to develop strong graduate instruction, nevertheless as umch as possible
should be done with the resources at hand to improve the training of the
young research workers who are on the staff; and beyond this, to give
them a chance to gp elsewhere for further training. 0n the whole, the
situation with resPect to opportunities for graduate study is not al-
together good, and the training of research workers is preceding too
slowly. ‘ ' '

3. Sabbatical leave: A survey of the state institutions
as to the matter of sabbatical leave in-
, dicates that a large majority of the‘in-
stitutions do not grant any such leave Whatsoever. Thirty-eight states
replied to a question on this subject. There are twentystwo which do
not have any provision for sabbatiCal leave. Five occasionally give'
leaves of absence, four of these without any definite plan. Twelve

466

grant sabbatical leave under certain conditions. There is apparently
no assurance, however, that a worker will be granted a leave of absence.
Usually the conditions are quite rigid, and the leave is granted only in
special cases with the consent of the board of trustees or the president
or both. In one state (Mass.), leave of absence is forbidden. In
another institution it is apparently pessible to get a leaye of absence
by working during the summer and in this way accumulate a sufficient
credit so that a leave of absence may be granted on full pay (Wis.).
This, however, cannot be considered a leave of absence inasmuch as the
person has already put in the time necessary for it. On the whole, the
situation with regard to sabbatical leave is not encouraging, nor is it
likely to become so, unless some effort is put forth to get the institus
tions to realize more fully the necessity for granting leave for further

study.

4. Term of service: The term of service imposed upon
’ research workers by most institutions is
excessive and on the whole is bad for
both the workers and the research agency.
For example, in 1926—27,-out of 288 workers engaged in economic research
in agriculture, 204, or 71 percent were employed on the eleven or twelve-
month basis. Employment on this basis restricts the worker very much,
sepecially with respect to obtaining additional training and experience.
If the worker wishes to pursue graduate studies or acquire additional
training, he must obtain leave of absence from the institution or resign
his position.. This is especially serious in view of the fact that a
large majority of the institutions in which research is being conducted
cannot provide resident graduate work of an outstanding character. Ebr-
thermors, it is often impossible for the workers to take graduate work
even if the institution is in a position to give such training; for_in
order to do so he is obliged to work on a part-time basis and thus re—
duce his salary Which is in most cases already too low.

Frequently there is discrimination against the research work
in favor of those who are engaged in teaching exclusively. In many
institutions the teaching-staff is on a nine-month basis. This gives
teachers ample opportunity to attend summer school, and thus over a .
period of years to obtain sufficient graduate credit to obtain their
advanced degree. Apparently, it has been recognized that in the case
of teachers, further study and training is necessary. Similar recogni-
tion should be given to research workers.

There is need also for a more definite plan of promotion of
research workers. Complaint is often heard among research workers in
this field that there is no opportunity for advancement. Provision for
this would not only increase the morale of the workers but also aid
in retaining the services of the young men with ability.